<!DOCTYPE html><html lang="en" xmlns="http://www.w3.org/1999/xhtml" xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" style="font-size:16px;"><head></head><head><meta charset="utf-8"/><!--[if !mso]><!--><meta http-equiv="X-UA-Compatible" content="IE=edge"/><!--<![endif]--><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="x-apple-disable-message-reformatting"/><meta name="format-detection" content="telephone=no,address=no,email=no,date=no,url=no"/><meta name="color-scheme" content="light"/><meta name="supported-color-schemes" content="light"/><title>Inductive Moment Matching</title><!--[if mso]><xml><o:OfficeDocumentSettings><o:AllowPNG/><o:PixelsPerInch>96</o:PixelsPerInch></o:OfficeDocumentSettings></xml><![endif]--><style>
:root { color-scheme: light; supported-color-schemes: light; }
body { margin: 0; padding: 0; min-width: 100%!important; -ms-text-size-adjust: 100% !important; -webkit-transform: scale(1) !important; -webkit-text-size-adjust: 100% !important; -webkit-font-smoothing: antialiased !important; }
.body { word-wrap: normal; word-spacing:normal; }
table.mso { width: 100%; border-collapse: collapse; padding: 0; table-layout: fixed; }
img { border: 0; outline: none; }
table { mso-table-lspace: 0px; mso-table-rspace: 0px; }
td, a, span { mso-line-height-rule: exactly; }
#root [x-apple-data-detectors=true],
a[x-apple-data-detectors=true],
#MessageViewBody a { color: inherit !important; text-decoration: inherit !important; font-size: inherit !important; font-family: inherit !important; font-weight: inherit !important; line-height: inherit !important; }
span.MsoHyperlink { color: inherit !important; mso-style-priority: 99 !important; }
span.MsoHyperlinkFollowed { color: inherit !important; mso-style-priority: 99 !important; }
.a { background-color:#dedede; }
.b { background-color:#2a2a2a; }
.c { background-color:#ffffff; }
.d { background-color:#fff0c8; }
.d2 { background-color:#FFFFFF; }
.d3 { background-color:#FFFFFF; }
h1 a { text-decoration:none;color:#2A2A2A !important; }
h2 a { text-decoration:none;color:#2A2A2A !important; }
h3 a { text-decoration:none;color:#2A2A2A !important; }
h4 a { text-decoration:none;color:#2A2A2A !important; }
h5 a { text-decoration:none;color:#2A2A2A !important; }
h6 a { text-decoration:none;color:#2A2A2A !important; }
h1, h1 a, h2, h2 a, h3, h3 a, h4, h4 a, h5, h5 a, h6, h6 a, ul, li, ol, p, p a { margin: 0;padding: 0; }
h1 { font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif;font-weight:700;font-size:28px;color:#2A2A2A;line-height:42px;padding-bottom:4px;padding-top:16px;mso-margin-top-alt:16px;mso-margin-bottom-alt:4px }
h2 { font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif;font-weight:700;font-size:24px;color:#2A2A2A;line-height:36px;padding-bottom:4px;padding-top:16px;mso-margin-top-alt:16px;mso-margin-bottom-alt:4px }
h3 { font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif;font-weight:400;font-size:20px;color:#2A2A2A;line-height:30px;padding-bottom:4px;padding-top:16px;mso-margin-top-alt:16px;mso-margin-bottom-alt:4px }
h4 { font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif;font-weight:400;font-size:18px;color:#2A2A2A;line-height:27px;padding-bottom:4px;padding-top:16px;mso-margin-top-alt:16px;mso-margin-bottom-alt:4px }
h5 { font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif;font-weight:400;font-size:16px;color:#2A2A2A;line-height:24px;padding-bottom:4px;padding-top:16px;mso-margin-top-alt:16px;mso-margin-bottom-alt:4px }
h6 { font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif;font-weight:400;font-size:14px;color:#2A2A2A;line-height:21px;padding-bottom:4px;padding-top:16px;mso-margin-top-alt:16px;mso-margin-bottom-alt:4px }
p { font-family:'Georgia','Times New Roman',serif;font-weight:400;color:#2D2D2D;font-size:16px;line-height:24px;padding-bottom:8px;padding-top:8px;mso-margin-top-alt:8px;mso-margin-bottom-alt:8px; }
p a, .e a, ul a, li a, .h a, .h2 a, .h3 a { word-break:break-word;color:#2C81E5 !important;text-decoration:none;font-style:italic; }
p a span, .e a span, ul a span, li a span { color: inherit }
p .bold { font-weight:bold;color:#2D2D2D; }
p span[style*="font-size"] { line-height: 1.6; }
.f p { font-size:12px;line-height:15px;color:#2D2D2D;padding:0; }
.f p a { color:#2D2D2D !important; }
.g p { font-family:'Helvetica',Arial,sans-serif;font-size:14px;line-height:20px;font-weight:normal;margin:0; }
.g p a { text-decoration: underline; }
.i p { font-family:'Helvetica',Arial,sans-serif;line-height:23px;font-size:15px;color:#2D2D2D; }
.i p a { color:#2D2D2D !important; }
.i2 p { font-family:'Helvetica',Arial,sans-serif;line-height:23px;font-size:15px;color:#2D2D2D; }
.i2 p a { color:#2D2D2D !important; }
.i3 p { font-family:'Helvetica',Arial,sans-serif;line-height:43px;font-size:24px;color:#2D2D2D; }
.i3 p a { color:#2D2D2D !important; }
.h p a { color:#595959 !important; }
.h2 p a { color:#595959 !important; }
.h3 p a { color:#595959 !important; }
.f p a, .i p a, .i2 p a, .i3 p a, .h p a, .h2 p a, .h3 p a { text-decoration:underline; }
.j { border-top:3px solid #ffeb2d; }
.k p { padding-left:15px;padding-bottom:0px;padding-top:6px;mso-margin-top-alt:6px;mso-margin-bottom-alt:0px;mso-margin-left-alt:15px; }
.o { background-color:#FFFFFF;border:1px solid #F1F1F1;border-radius:5px; }
.o p { font-family:'Helvetica',Arial,sans-serif;padding:0px;margin:0px; }
.l p,
.l p a { font-size:14px;line-height:20px;font-weight: bold;color:#2D2D2D;padding-bottom:6px;mso-margin-bottom-alt:6px;text-decoration:none; }
.m p,
.m p a { font-size:13px;line-height:18px;font-weight:400;color:#2D2D2D;padding-bottom:6px;mso-margin-bottom-alt:6px;text-decoration:none; }
.n p,
.n p a { font-size:12px;line-height:17px;font-weight:400;color:#2D2D2D;padding-bottom:6px;mso-margin-bottom-alt:6px;text-decoration:none; }
.p { background-color:#FFFFFF;max-width:520px;border:1px solid #E1E8ED;border:1px solid rgba(80, 80, 80, 0.3);border-radius:5px; }
.q { font-size:16px;font-family:Helvetica,Roboto,Calibri,sans-serif !important;border:1px solid #e1e8ed;border:1px solid rgba(80, 80, 80, 0.3);border-radius:10px;background-color:#FFFFFF; }
.q p { font-size:16px;font-family:system-ui,Helvetica,Roboto,Calibri,sans-serif !important;color:#222222;padding:4px 0; }
.r { border:1px solid #E1E8ED !important;border-radius:5px; }
.s p { font-size: 14px; line-height: 17px; font-weight: 400; color: #697882; text-decoration: none; }
.t p { font-family:'Helvetica',Arial,sans-serif;font-size:12px;line-height:18px;font-weight:400;color:#000000;font-style:italic;padding:4px 0px 0px;}
.v { border-radius:10px;border:solid 0px #DFD150;background-color:#2C81E5;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;color:#FFFFFF; }
.v a { text-decoration:none;display:block;color:#FFFFFF; }
.w p { font-size:12px;line-height:15px;font-weight:400;color:#FFFFFF; }
.w p a { text-decoration: underline !important;color:#FFFFFF !important; }
ul { font-family:'Helvetica',Arial,sans-serif;margin:0px 0px 0px 25px !important;padding:0px !important;color:#2D2D2D;line-height:24px;list-style:disc;font-size:16px; }
ul > li { font-family:'Helvetica',Arial,sans-serif;margin:10px 0px 0px 0px !important;padding: 0px 0px 0px 0px !important; color: #2D2D2D; list-style:disc; }
ol { font-family:'Helvetica',Arial,sans-serif;margin: 0px 0px 0px 25px !important;padding:0px !important;color:#2D2D2D;line-height:24px;list-style:decimal;font-size:16px; }
ol > li { font-family:'Helvetica',Arial,sans-serif;margin:10px 0px 0px 0px !important;padding: 0px 0px 0px 0px !important; color: #2D2D2D; list-style:decimal; }
.e h3,
.e p,
.e span { padding-bottom:0px;padding-top:0px;mso-margin-top-alt:0px;mso-margin-bottom-alt:0px; }
.e span,
.e li { font-family:'Helvetica',Arial,sans-serif;font-size:16px;color:#2D2D2D;line-height:24px; }
.rec { font-family: ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol", "Noto Color Emoji" !important; }
.rec__button:hover { background-color: #f9fafb !important; }
.copyright a {color: inherit !important; text-decoration: none !important; font-size: inherit !important; font-family: inherit !important; font-weight: inherit !important; line-height: inherit !important;}
.txt_social p { padding: 0; word-break: break-all; }
.table, .table-c, .table-h { border: 1px solid #C0C0C0; }
.table-c { padding:5px; background-color:#FFFFFF; }
.table-c p { color: #2D2D2D; font-family:'Helvetica',Arial,sans-serif !important;overflow-wrap: break-word; }
.table-h { padding:5px; background-color:#F1F1F1; }
.table-h p { color: #2A2A2A; font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif !important;overflow-wrap: break-word; }
@media only screen and (max-width:667px) {
.aa { width: 100% !important; }
.bb img { width: 100% !important; height: auto !important; max-width: none !important; }
.cc { padding: 0px 8px !important; }
.ee { padding-top:10px !important;padding-bottom:10px !important; }
.ff ul, .ff ol { margin: 0px 0px 0px 10px !important;padding: 0px !important; }
.ff li { margin:10px 0px 0px 10px !important; }
.r {height:140px !important;}
.s p { font-size:13px !important;line-height:15px !important; }
.mob-hide {display:none !important;}
.mob-stack {display:block !important;width:100% !important;}
.mob-w-full {width:100% !important;}
.mob-block {display:block !important;}
.embed-img {padding:0px 0px 12px 0px !important;}
.socialShare {padding-top:15px !important;}
.rec { padding-left:15px!important;padding-right:15px!important; }
.bodyWrapper { padding:7px 4px 7px 4px !important; }
.social-mobile {float:left !important;margin-top:10px !important;}
}
@media screen and (max-width: 480px) {
u + .a .gg { width: 100% !important; width: 100vw !important; }
.tok-heart { padding-top:75% !important; }
.tok-play { padding-top: 250px !important; }
}
@media screen and (max-width: 320px) {
.tok-heart { padding-top:65% !important; }
}
.u { border: 1px solid #CACACA !important; border-radius: 2px !important; background-color: #ffffff !important; padding: 0px 13px 0px 13px !important; font-family:ui-sans-serif,system-ui,-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,"Noto Sans",sans-serif !important;font-size: 12px !important; color: #767676 !important; }
.u a { text-decoration: none; display: block !important; color: #767676 !important; margin: 0px !important; }
.u span, .u img { color: #767676 !important;margin:0px !important; max-height:32px !important;background-color:#ffffff !important; }
</style><!--[if mso]><style type="text/css">
sup { font-size: 100% !important;vertical-align: .5em !important;mso-text-raise: -1.5% !important;line-height: 0 !important; }
ul { margin-left:0px !important; margin-right:10px !important; margin-top:20px !important; margin-bottom:20px !important; }
ul li { margin-left: 0px !important; mso-special-format: decimal; }
ol { margin-left:0px !important; margin-right:10px !important; margin-top:20px !important; margin-bottom:20px !important; }
ol li { margin-left: 0px !important; mso-special-format: decimal; }
li.listItem { margin-left:15px !important; margin-top:0px !important; }
.paddingDesktop { padding: 10px 0 !important; }
.edm_outlooklist { margin-left: -20px !important; }
.embedImage { display:none !important; }
</style><![endif]--><style>
@font-face {
font-family: 'Open Sans';
font-style: normal;
font-weight: 700;
font-display: swap;
src: url('https://fonts.gstatic.com/s/opensans/v40/memSYaGs126MiZpBA-UvWbX2vVnXBbObj2OVZyOOSr4dVJWUgsg-1x4gaVIUwaEQbjA.woff2') format('woff2');
}
@font-face {
font-family: 'Open Sans';
font-style: italic;
font-weight: 700;
font-display: swap;
src: url('https://fonts.googleapis.com/css2?family=Open+Sans:ital,wght@1,700&display=swap') format('woff2');
}
</style></head><body class="a" style="margin:0px auto;padding:0px;word-wrap:normal;word-spacing:normal;background-color:#dedede;"><div role="article" aria-roledescription="email" aria-label="email_name" lang="en" style="font-size:1rem"><div style="display:none;max-height:0px;overflow:hidden;"> Plus more about Generalized Kullback-Leibler Divergence Loss and Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ </div><table role="none" width="100%" border="0" cellspacing="0" align="center" cellpadding="0" class="gg"><tr><td align="center" valign="top"><table role="none" width="670" border="0" cellspacing="0" cellpadding="0" class="aa" style="width:670px;table-layout:fixed;"><tr><td class="bodyWrapper" align="center" valign="top" style="padding:7px 7px 7px 7px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td align="center" valign="top" style="border:4px solid #2a2a2a;border-radius:0px;background-color:#ffffff;" class="c"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td class="f" align="right" valign="top" style="padding:20px 28px;"><p> March 19, 2025 | <a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.c6q0w4g5sodbtO4I1B_pxSdB5RCIH6yy1Fm1CYma3Ez33iqeLmsDpcnAqaMynMgc9dSeusqrsJkgq_ohmE7Dqgy4hrkyUCxmrha7A0fKwTTqViYKm78szw_r4CfUtNJKkFX9VldNZ0_EtnbmfyYMavkJr6XK-3RYIqqa4V5VsgXz6lS4uFnPktGZ77LIVgjELOeAAndt3hnO0XqV4xnbiD5Xu9quxIHrD9OJLbg0CAN96QnCd43mRPgwrammQvm2WBQRtoo9uE9-bAnFIR4eAD3swKmPGq335r7rUU8mAQBmLes9gTH4L3hizMUcyofgeD9sV0xEjFMQU5V45BM5nXSMyr7WKm7GFko75BiisajyFm8OQswP237b7mY5MQItLH5hKcabqhj3Lys2mMXTL5HhP-yNQ5sIq2SuHIBvAjH9W6Riyj623eciQaiwHDpbCastzbflFS3uuThTtvg__8xS_W0REihg-_7ytJVSckSioS2h-bmBDJByMY2YXJvaWAjlWsz5SWxGPEw-I-S-x4qox6hxF4Qcdy6cGMq3v58NcCfuOn0tUyu4j_JKPLQtgsdK967GLLXUeso1QOfItcWwyhV6qJsApmrO__TDs67qYYYaJPgQl5ItyW9bZOWrGOh1Brwd0-7L9DzpDqm2D6xhNdEzvEWSra3FI8c-UFQ/4ex/Or5yX8RKQt6FNRtIx1tHyg/h0/h001.JKlrpEaTQhNlmJad22lUoCgyjU_bia0fb-Q1I3oebjY">Read Online</a></p></td></tr><tr><td class="dd" align="center" valign="top" style="padding:15px 28px 20px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td align="center" valign="top"><h1 style="text-align:left;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;font-weight:Bold;font-size:32px;color:#2A2A2A;padding:2px 0;line-height:38px;"> Inductive Moment Matching </h1><p style="text-align:left;font-family:'Helvetica',Arial,sans-serif;font-weight:normal;font-size:20px;color:#3E3E3E;padding:5px 0;line-height:24px;"> Plus more about Generalized Kullback-Leibler Divergence Loss and Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models </p></td></tr></table></td></tr><tr><td style="height:0px;width:0px;"><div style="height:1px;" data-open-tracking="true"> <img src="https://elink4f7.mail.bycloud.ai/ss/o/u001.3wmUuY8gEWd4_869a_eXcg/4ex/Or5yX8RKQt6FNRtIx1tHyg/ho.gif" alt="" width="1" height="1" border="0" style="height:1px !important;width:1px !important;border-width:0 !important;margin-top:0 !important;margin-bottom:0 !important;margin-right:0 !important;margin-left:0 !important;padding-top:0 !important;padding-bottom:0 !important;padding-right:0 !important;padding-left:0 !important;"/> </div></td></tr><tr id="content-blocks"><td class="email-card-body" align="center" valign="top" style="padding-bottom:28px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td id="nov-18-th-nov-24-th-33-latest-ai-re" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h6 style="color:#2A2A2A;font-weight:normal;"><i>Mar 10th ~ Mar 17th</i><br><i>#47 Latest AI Research Explained Simply</i></h6></td></tr><tr><td><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" style=""><tr><td bgcolor="#222222" style="background-color:#222222;padding:0.0px 0.0px 0.0px 0.0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0"><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"></p></td></tr></table></td></tr></table></td></tr><tr><td id="industry-news-in-1-line" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:Bold;padding:0px 28px;text-align:left;"><h2 style="color:#2A2A2A;font-weight:Bold;">🗞️ Industry News in 1 Line</h2></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="background-color:#e0e0e0;"><span style="color:rgb(255, 58, 58);font-size:0.6rem;">♥ 7.2k</span></span> <a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.c6q0w4g5sodbtO4I1B_pxU6myLBNr0kxu76gma1gUlZByUWSSlQT-aMnpvCwPB55zDOItff8u11B1d0DJjRAhmCoEODLpnibeTfWaOvJYatkOj6tg3E2k3gsGEOE0WOoYv47ygRT6X6uqa7AKrX9YX4e_gNalgkuNMQv-AqNr0A/4ex/Or5yX8RKQt6FNRtIx1tHyg/h1/h001.S2-1EfRD4XGwswfzdlomUb6HiADqmSl3ggb8XkFQV64" target="_blank" rel="noopener noreferrer nofollow"><span>Mistral Small 3.1</span></a> is a fast, lightweight multimodal model with a 128k token context window, outperforming Gemma 3 and GPT-4o Mini while running efficiently on consumer hardware with 32GB VRAM. Both <a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.CxDkkVpJsBdVoe83c_tBWiKss6PuQ-sZmR6_6eHSgi9XG9ls1Fr7cId3z0KX13rXt6Y0lv9EF_6mOVwYuHxFtjxkjggQaSOrnmrT7i4BxIpqDHxjY6GQKhQJDuvWlqgnM41LthOj-zzga8wRTVTYA2o4_AQ-gAACkmdmsPMdbQo/4ex/Or5yX8RKQt6FNRtIx1tHyg/h2/h001.c_SGADdzID9b68QAYenfjVsXEjB-QtIt08w_KCmO5_Y" target="_blank" rel="noopener noreferrer nofollow"><span>base</span></a> and <a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.CxDkkVpJsBdVoe83c_tBWiKss6PuQ-sZmR6_6eHSgi9XG9ls1Fr7cId3z0KX13rXr6etvKgsjYgDfrq-lkK8XiI1ZSAUxYMUnB1R_ItnGE0vnCDhxSJISLlA5kptVppLtJWj1QcEh4YM2r0Q1999Dlev0KOZjkXe5nsHuecJmAc/4ex/Or5yX8RKQt6FNRtIx1tHyg/h3/h001.eF8EJGK2348Rpui7e2cJAzXvFXbsOZsxcbUf0ZnYeLQ" target="_blank" rel="noopener noreferrer nofollow"><span>instruct</span></a> model are now available on HuggingFace. </p><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:480px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/4765ae63-3be2-4782-b13d-f7935f253469/Screenshot_2025-03-18_at_8.13.27_PM.png?t=1742354013" alt="" height="auto" width="480" style="display:block;width:100%;" border="0"/></td></tr></table></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="background-color:#e0e0e0;"><span style="color:rgb(255, 58, 58);font-size:0.6rem;">♥ 4.5k</span></span> <a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.1muhFWIqieRYpaJ-FbWSCdZre9zuaXlYXS8KZVZW1Jz40iV452p7_QE1ylORXWHTgnyaBIvlJe1uZ40RMXFIhdhaTLLc_2XV5lDwF-bCXt7jMnU8oRbrZt9xoIrxUTcV2nln0NaGJeXbDxNlIzuv7A/4ex/Or5yX8RKQt6FNRtIx1tHyg/h4/h001.LZiDvS7yeTl5hIe_kI505XnhRsATgRkThoKqbg3yrgg" target="_blank" rel="noopener noreferrer nofollow"><span>ERNIE 4.5, a next-gen multimodal foundation model, and ERNIE X1</span></a>, a deep-thinking reasoning model, now offer top-tier performance at lower costs, with ERNIE X1 matching DeepSeek R1 at half the price. ERNIE 4.5 excels in reasoning, memory, and hallucination prevention, while both models are freely available via <a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.amatuKKICSickUKplYJXmBxNZR4lWEL_cT87t-jPwXkbXQqwIytofA7QAmQRmTJ10TZelDpYrS0gMEoK-YlGzMmemHYMEmBx5azy_3SrxbloyfAEL6_B-02UZ499B_uT9EuY41FSHkLJlJNcL_megg/4ex/Or5yX8RKQt6FNRtIx1tHyg/h5/h001.Nq8SaXsWMMrex-LNbKFub-4PxXr6JC8_6nXq-CPW2zI" target="_blank" rel="noopener noreferrer nofollow"><span>ERNIE Bot</span></a> and accessible to enterprises through Baidu AI Cloud. </p></li></ol></div></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:330px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/d0998083-3c6d-4d7c-bf99-bdc3c0e5b3ed/GmIKgSobcAAMR_W.jpeg?t=1742354577" alt="" height="auto" width="330" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:330px;"><p>ERNIE 4.5 benchmark</p></td></tr></table></td></tr><tr><td><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" style=""><tr><td bgcolor="#222222" style="background-color:#222222;padding:0.0px 0.0px 0.0px 0.0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0"><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"></p></td></tr></table></td></tr></table></td></tr><tr><td><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" style=""><tr><td bgcolor="transparent" style="background-color:transparent;border-color:#2C81E5;border-style:solid;border-width:5px;padding:0.0px 0.0px 0.0px 0.0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0"><tr><td class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:Bold;padding:0px 28px;text-align:left;"><h2 style="color:#2A2A2A;font-weight:Bold;"><span style="">RTX 4080 SUPER Giveaway (RIGHT NOW!) With NVIDIA’s GTC 2025</span></h2></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.1r-Oo1v2APbJapoJieKk3V2DO7NkBsza57dliEOwa6PrCHfGqwkzFzXNcPKf-_zR2eofoRDJVIr2uWPCL0O09M38ctdoxt0-4uft8xTqPF4A2L5bfiFSy8hNnpelPJkW/4ex/Or5yX8RKQt6FNRtIx1tHyg/h6/h001.AMzx-ZTAqVVGtcre9ZKADFROkZmb0u-a9tlHfDiy0ls" rel="noopener noreferrer nofollow" style="text-decoration:none;" target="_blank"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/6b4027f1-7cb0-483c-bc62-f7c3e56d1d36/geforce-rtx-4080-super-og-1200x630.jpg?t=1739758485" alt="RTX4080 SUPER Giveaway" height="auto" width="600" style="display:block;width:100%;border-radius:0px 0px 0px 0px;border-style:solid;border-width:0px 0px 0px 0px;box-sizing:border-box;border-color:#E5E7EB;" border="0"/></a></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p><span style="">RTX 4080 SUPER Giveaway!</span></p></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="">During NVIDIA’s GTC event which is NVIDIA’s annual flagship AI & developer Conference, </span><span style=""><b>March 17-21, 2025</b></span><span style="">, there will be various big announcements, events, and sessions you can </span><span style=""><b>attend both in-person or virtually</b></span><span style="">. </span></p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="">You can virtually discover the latest breakthroughs in generative AI and NVIDIA technologies from subject matter experts at </span><span style=""><b>#GTC25</b></span></p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.sa7HIrCkEzyny8lstY55mOGXqey6XwqgGMAgXuOux3xG-dEbFgoBohMjdiScpbE6OarznDTReYOnQFUI9Bc1j38TyGFk4-c7VfnSuhABE8Q/4ex/Or5yX8RKQt6FNRtIx1tHyg/h7/h001.yB7DdjDt2fudFCwrS8iILNm0dEcUDVXc05uxk6MntAk" rel="noopener noreferrer nofollow" style="text-decoration:none;" target="_blank"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/3ed23905-cca4-45e7-9d22-597420c6671a/Screenshot_2025-02-16_210612.png?t=1739757983" alt="Highlighted Technical Speakers List from GTC2025" height="auto" width="600" style="display:block;width:100%;border-radius:0px 0px 0px 0px;border-style:solid;border-width:0px 0px 0px 0px;box-sizing:border-box;border-color:#E5E7EB;" border="0"/></a></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p><span style="">Highlighted Technical Speakers List from GTC2025</span></p></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="">By virtually attending sessions, you can join my giveaway for an RTX4080 SUPER. </span><span style=""><b>Currently only 25 people has joined the giveaway.</b></span></p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="">All you have to do is to take a selfie of yourself attending the </span><span style=""><a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.sa7HIrCkEzyny8lstY55mOGXqey6XwqgGMAgXuOux3yu6qrZISwQvLbW-sBEWRxxbJysVXf9yVxXsRD_FkUkvU7_scgDqhPuFfdBQW-fkUg/4ex/Or5yX8RKQt6FNRtIx1tHyg/h8/h001.1vDQZtADaienFnc_AR9OH03SPjeeWX4e5tNaeKtcj0w" target="_blank" rel="noopener noreferrer nofollow"><span>LIVE virtual sessions</span></a></span><span style=""> that are available during GTC (March 17-21), submit it to </span><span style=""><a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.9ggl6Mt0xphuuMReR5gVpdhff8iVAkayl-q5vnwXNgVweea5z7E-FAjaLRloITHq8yyEUjt04McVD-wRtPzdBiM1tfVTT0WaI0pZkebK9Gwmmt6M0mcypod2k8OqurmIy8X8Gjtp9CgU3FqMla57lG88BbsqIbOWrVWNcK80lWc_x3ztksmyFHrD1Lx1KLG5W_ek9F96B5Uerrm3BHgyuw/4ex/Or5yX8RKQt6FNRtIx1tHyg/h9/h001.L7KEXy8HAnyvEbvZ0SW4YdkgDKh5C0dXOYB2PQCWVNA" target="_blank" rel="noopener noreferrer nofollow"><span>this Google Form</span></a></span><span style="">, and you can learn while possibly win a GPU at the same time! You can find more information on the google form.</span></p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/9d944eec-6f33-48a2-96ff-d69b1ab7d99c/Screenshot_2025-03-18_at_10.24.01_AM.png?t=1742318648" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p><span style="">GTC2025 Keynote by Jensen Huang</span></p></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="">Here is a </span><span style=""><b>summary of the Keynote</b></span><span style="">:</span></p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ul style="list-style-type:disc;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">A NVIDIA Dynamo infrastructure software that boosts 30x throughput on Blackwell</span></p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">A New Blackwell Ultra has 1.5x higher inference speed than Blackwell</span></p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">DGX Spark, the smallest AI supercomputer, has a 128GB unified memory</span></p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">RTX Pro, a new GPU series for consumers, which is using Blackwell architecture, ranges from 24GB up to 96GB</span></p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">Enterprise Agentic AI for all levels, with focus on reasoning for agents</span></p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">Llama Nemotron Reasoning Models, trained specifically for reasoning agentic use, has 3 sizes: Nano, Super, Ultra</span></p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="">AI-Q, an NVIDIA AI Blueprint that connects reasoning to AI agents, data, and tools for enterprise to connect data and deploy AI efficiently. </span></p></li></ul></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="">plus many other physical AI and simulation announcements, you can check out the </span><span style=""><a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.amatuKKICSickUKplYJXmGpeQ0d0i5QgmxJe9jWkSg0jLSskImtX5t1pmg0qzvaTQo6CkwU9_6CWPVtrTJqBmKk3uP4z0efivm3Q3gkuwZf7Tw5tZmjapj6ER2zziZRV/4ex/Or5yX8RKQt6FNRtIx1tHyg/h10/h001.z8x86m_PaEK_tcUvHHJCjDQtoN8NrgNnhWh2zLrB2AU" target="_blank" rel="noopener noreferrer nofollow"><span>GTC2025 Keynote replay on YouTube</span></a></span><span style="">.</span></p></td></tr><tr><td align="center" valign="top" style="padding-bottom:14px;padding-left:28px;padding-right:28px;padding-top:14px;text-align:center;width:100%;word-break:break-word;" class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" align="center" style=""><tr><td align="center" valign="middle" height="42" style="height:42px;background-color:#2C81E5;border-radius:10px;border:solid 0px #DFD150;color:#FFFFFF;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.sa7HIrCkEzyny8lstY55mOGXqey6XwqgGMAgXuOux3w2CUd6LM9zuUwC5FWxSXUBUIlQRYANMKEmcDHyQR6jQ31X9n8K6CxvRjhb2rCc4Pg/4ex/Or5yX8RKQt6FNRtIx1tHyg/h11/h001.btZXsGuLBZf0DfqaehXTmV0KvhH5HmxM0D4w6YtHDCw" target="_blank" rel="noopener noreferrer nofollow" style="color:#FFFFFF;display:block;font-size:16px;padding:0px 14px;text-decoration:none;"> Check Out GTC 2025! </a></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style=""><a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.tLfGW26lAwaS9gFg17HSoGymQ3NNPtd5dE5MV_8UgjIDFPVXngz8pvQBldSW42yhUe_Qiq6DgEPMEBuPL9yfRpXelTiuu2kS8pLFvsoem_XoZoy_n13sTKUhZIbl0VH6/4ex/Or5yX8RKQt6FNRtIx1tHyg/h12/h001.uvLxwKD3ZvjALWpQytjS-Td4fJwNeF0SFo6soBbN6yY" target="_blank" rel="noopener noreferrer nofollow"><span>Advertise with The AI Timeline! </span></a></span></p></td></tr></table></td></tr></table></td></tr><tr><td><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" style=""><tr><td bgcolor="#222222" style="background-color:#222222;padding:0.0px 0.0px 0.0px 0.0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0"><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"></p></td></tr></table></td></tr></table></td></tr><tr><td id="inductive-moment-matching" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:Bold;padding:0px 28px;text-align:left;"><h2 style="color:#2A2A2A;font-weight:Bold;">Inductive Moment Matching</h2></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style=""><i>Zhou et al. [Luma AI, Stanford University]</i></span></p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="background-color:#e0e0e0;"><span style="color:rgb(255, 58, 58);font-size:0.6rem;"> ♥ 2.6k </span></span><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> </span><span style="background-color:#e0e0e0;"><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> Diffusion Models </span></span></p></td></tr><tr><td id="introduction-to-inductive-moment-ma" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Introduction to Inductive Moment Matching</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> There are many challenging problems in the field of generative AI: achieving high-quality outputs, efficient inference, and stable training simultaneously. Current approaches like diffusion models produce impressive results but require many inference steps, while attempts to accelerate them through distillation or Consistency Models often lead to instability and extensive hyperparameter tuning. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> This paper aims to solve this by introducing Inductive Moment Matching (IMM), a single-stage training procedure that directly learns generative models capable of high-quality one-step or few-step sampling without requiring pre-training or model distillation. By operating on time-dependent marginal distributions and enforcing distribution matching through mathematical induction, IMM guarantees convergence to the data distribution while maintaining stability across various hyperparameters and model architectures. </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/fc04b365-f765-49dd-b5d2-0b3aa70a6a8f/image.png?t=1742313294" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr></table></td></tr><tr><td id="how-inductive-moment-matching-works" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">How Inductive Moment Matching Works</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Inductive Moment Matching (IMM) creates AI-generated images in just a few steps instead of hundreds, while maintaining high quality and training stability. IMM uses a clever shortcut - directly learning how to transform noise into images without the lengthy step-by-step process traditional models use. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="font-weight:700;"><b>The Interpolation Framework</b></span></p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ul style="list-style-type:disc;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> IMM creates a continuous path between random noise (t=1) and real images (t=0) </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> At any point in time t, there exists a distribution of partially-noised images </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> The key insight: learn to jump directly from any time point to any earlier time point </p></li></ul></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="font-weight:700;"><b>Learning Through Induction</b></span></p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ul style="list-style-type:disc;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> IMM learns by comparing two routes to the same destination: </p><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> Going directly from time t to time s </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> Going from t to an intermediate time r, then from r to s </p></li></ol></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> When these two routes produce the same results, the model has learned correctly </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> This "inductive bootstrapping" technique guarantees the model converges to the correct distribution </p></li></ul></div></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/29a2d3e5-eaae-4833-aa31-21f293281dde/image.png?t=1742313323" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="font-weight:700;"><b>Training Stability</b></span></p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ul style="list-style-type:disc;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> IMM uses "moment matching" (comparing statistical properties between distributions) </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> Instead of matching single samples, it matches multiple "particles" at once </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> This makes training much more stable than previous approaches </p></li></ul></div></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/fe9d2aef-9035-461d-b6a0-67d7476a26e8/image.png?t=1742313352" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr></table></td></tr><tr><td id="results-and-real-world-implications" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Results and Real-World Implications of Inductive Moment Matching</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Researchers tested different approaches to find optimal configurations for Inductive Moment Matching. Across architectures (DDPM++ for CIFAR-10 and DiT-B for ImageNet-256×256), they found that network parameterization and flow schedules <span style="font-weight:700;"><b>significantly impact performance</b></span>. While Simple-EDM with OT-FM flow excels on smaller datasets, the Euler-FM combination demonstrates superior scalability on larger images, which suggests different optimal configurations based on resolution. </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/9d349b11-498c-41f2-946d-95131de7669f/image.png?t=1742313414" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Researchers analyzed different mapping functions and concluded that constant decrements in η_t consistently outperform alternative approaches, though stability requires careful selection of decrements proportional to 2^-k. Weighting functions also proved crucial, with the combination of ELBO factors, α_t weighting, and middle time-step emphasis via 1/(α²_t+σ²_t) yielding substantial improvements. </p></td></tr><tr><td align="center" valign="top" style="padding-bottom:14px;padding-left:28px;padding-right:28px;padding-top:14px;text-align:center;width:100%;word-break:break-word;" class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" align="center" style=""><tr><td align="center" valign="middle" height="42" style="height:42px;background-color:#2C81E5;border-radius:10px;border:solid 0px #DFD150;color:#FFFFFF;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.DUiN96-Eq7pUHzwEhy5j28yjf9KIXZdsXoh1WlHvvKkyP1nEqKnihHzkFQfHlVHQEW5Iqfs8d69pYWTU2tGqMuy8V33VBaefb50KYi21d5qV5S3Mp6goC8x71e3vwSHc/4ex/Or5yX8RKQt6FNRtIx1tHyg/h13/h001.NlhvcldgCVrwGGyw9YAmM5cNT6wIn3mzz5wi0CY6_RU" target="_blank" rel="noopener noreferrer nofollow" style="color:#FFFFFF;display:block;font-size:16px;padding:0px 14px;text-decoration:none;"> Read Full Paper </a></td></tr></table></td></tr><tr><td><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" style=""><tr><td bgcolor="#222222" style="background-color:#222222;padding:0.0px 0.0px 0.0px 0.0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0"><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"></p></td></tr></table></td></tr></table></td></tr><tr><td id="generalized-kullback-leibler-diverg" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:Bold;padding:0px 28px;text-align:left;"><h2 style="color:#2A2A2A;font-weight:Bold;">Generalized Kullback-Leibler Divergence Loss</h2></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style=""><i>Cui et al. [Nanyang Technological University, The Chinese University of Hong Kong, The University of Hong Kong, Harbin Institution of Technology, Hefei University of Technology]</i></span></p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="background-color:#e0e0e0;"><span style="color:rgb(255, 58, 58);font-size:0.6rem;"> ♥ 312 </span></span><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> </span><span style="background-color:#e0e0e0;"><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> KL Divergence </span></span><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> </span></p></td></tr><tr><td id="drawbacks-and-limitations-of-kullba" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Drawbacks and Limitations of Kullback-Leibler Divergence Loss </span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> This paper addresses limitations in the Kullback-Leibler (KL) Divergence loss function widely used in deep learning. The authors mathematically prove that KL loss can be decoupled into two components: a weighted Mean Square Error and Cross Entropy with soft labels. This decoupling reveals two critical weaknesses: asymmetric optimization that hinders convergence during knowledge distillation, and sample-wise prediction bias. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> To address this problem, they proposed a solution, Generalized Kullback-Leibler (GKL) Divergence loss, which breaks this asymmetric property while introducing a smoother weight function and class-wise global information. The effectiveness of GKL is demonstrated through impressive experimental results, achieving state-of-the-art adversarial robustness on RobustBench and competitive knowledge distillation performance across CIFAR, ImageNet, and CLIP models. </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/bea7440b-a175-452d-b3a0-997786bee5ff/image.png?t=1742313680" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p>Comparisons of gradient backpropagation between KL, DKL, and GKL losses.</p></td></tr></table></td></tr><tr><td id="understanding-kl-loss-decoupling" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Understanding KL Loss Decoupling</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The authors of this paper argue that Kullback-Leibler (KL) Divergence loss, widely used in deep learning, can be decoupled into two simpler components: </p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Weighted Mean Square Error (wMSE)</b></span> - Captures local relationships between pairs of classes </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Cross-Entropy with soft labels</b></span> - Ensures global similarity between probability distributions </p></li></ol></div></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/46056437-9824-4bb2-9732-0e28c414d03a/image.png?t=1742313777" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> This decoupling exposes two key problems in KL loss: </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="font-weight:700;"><b>Asymmetric Optimization Issue</b></span>: In knowledge distillation, where a student model learns from a fixed teacher model, the wMSE component becomes ineffective because the teacher's outputs are detached from gradient calculations. This means only half of the optimization mechanism works, leading to worse performance. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="font-weight:700;"><b>Sample-wise Prediction Bias</b></span>: Hard examples or outliers with incorrect predictions can mislead the training process when using sample-based weights. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The authors' solution (GKL loss) addresses these issues by: </p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> Breaking the asymmetric property - Allowing gradients to flow through both components </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> Introducing a smoother weight function - Making training more stable for classes with high predicted scores </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"> Using class-wise information - Reducing bias from individual sample predictions by incorporating global class statistics </p></li></ol></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> These improvements significantly enhance performance in both adversarial training and knowledge distillation without requiring mathematical reformulation of the original loss function. </p></td></tr><tr><td id="results-and-evaluation-of-kullback-" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Results and Evaluation of Kullback-Leibler Divergence Loss</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The authors' Generalized Kullback-Leibler (GKL) Divergence loss achieved <span style="font-weight:700;"><b>state-of-the-art adversarial robustness</b></span> on CIFAR-10/100 and delivered competitive knowledge distillation performance across CIFAR-10/100, ImageNet, and CLIP models. By decoupling KL loss into weighted MSE and Cross-Entropy components, breaking asymmetric optimization, and incorporating class-wise global information with smoother weight functions, GKL-AT significantly outperformed baselines across various attacks and perturbation sizes, demonstrating 1.34% higher average robustness than TRADES. </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/dcda2f7f-abc9-41e4-a548-c4e282464ee1/image.png?t=1742314160" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> However, despite these advances, the authors acknowledge GKL's effectiveness has only been demonstrated in adversarial training and knowledge distillation, suggesting future work should explore its applications in out-of-distribution robustness and incremental learning scenarios. </p></td></tr><tr><td align="center" valign="top" style="padding-bottom:14px;padding-left:28px;padding-right:28px;padding-top:14px;text-align:center;width:100%;word-break:break-word;" class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" align="center" style=""><tr><td align="center" valign="middle" height="42" style="height:42px;background-color:#2C81E5;border-radius:10px;border:solid 0px #DFD150;color:#FFFFFF;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.DUiN96-Eq7pUHzwEhy5j28yjf9KIXZdsXoh1WlHvvKlcGHuv-rTYMKX23ZvKZQ3Ey-IQ-9oMrjz4jYiOfoJQf0al6nJWvp4ymCU3KNa-K45niSQyZLufUYvIMCjrbagW/4ex/Or5yX8RKQt6FNRtIx1tHyg/h14/h001.epkIrDcnoyO8qyc7h1TrpM3t6DtrnwYuVD-I45IAGcU" target="_blank" rel="noopener noreferrer nofollow" style="color:#FFFFFF;display:block;font-size:16px;padding:0px 14px;text-decoration:none;"> Read Full Paper </a></td></tr></table></td></tr><tr><td><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" style=""><tr><td bgcolor="#222222" style="background-color:#222222;padding:0.0px 0.0px 0.0px 0.0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0"><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"></p></td></tr></table></td></tr></table></td></tr><tr><td id="vision-r-1-incentivizing-reasoning-" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:Bold;padding:0px 28px;text-align:left;"><h2 style="color:#2A2A2A;font-weight:Bold;">Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models</h2></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style=""><i>Huang et al. [East China Normal University, Xiaohongshu Inc.]</i></span></p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"><span style="background-color:#e0e0e0;"><span style="color:rgb(255, 58, 58);font-size:0.6rem;"> ♥ 568 </span></span><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> </span><span style="background-color:#e0e0e0;"><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> Vision Reasoning </span></span><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> </span><span style="background-color:#e0e0e0;"><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> bycloud’s pick </span></span><span style="color:rgb(44, 129, 229);font-size:0.6rem;"> </span></p></td></tr><tr><td id="introduction-to-vision-r-1" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Introduction to Vision-R1</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> This paper highlights the challenge of creating Multimodal Large Language Models (MLLMs) with robust reasoning capabilities, similar to those recently demonstrated in LLMs using Reinforcement Learning (RL) like DeepSeek-R1. One of the biggest problems is the scarcity of high-quality, multimodal data that includes complex, human-like reasoning steps (Chain-of-Thought or CoT), as opposed to simpler "Pseudo-CoT" data. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> To tackle this, the researchers introduce Vision-R1, a novel reasoning MLLM. They first create a large (200K) multimodal CoT dataset, Vision-R1-cold, without manual annotation. This is achieved by an innovative "Modality Bridging" technique where an existing MLLM generates initial "Pseudo-CoT" data, which is then refined using a powerful text-based reasoning LLM (DeepSeek-R1) and filtering. This dataset provides a "cold-start" for Vision-R1. </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Then, to overcome the "overthinking" issue observed during RL training, they propose Progressive Thinking Suppression Training (PTST) combined with Group Relative Policy Optimization (GRPO) and a specialized reward function. This progressively shapes the model's reasoning on a smaller (10K) multimodal math dataset, leading to a significant average improvement of ~6% on various benchmarks, with Vision-R1-7B achieving near state-of-the-art performance (73.5% on MathVista, close to OpenAI's O1). </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/7fe78838-f0fb-4b69-896b-d3d529640b21/image.png?t=1742314204" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p>Vision-R1 Pipeline</p></td></tr></table></td></tr><tr><td id="can-we-teach-reasoning-to-mll-ms-ju" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Can we teach reasoning to MLLMs </span><span style="color:rgb(67, 67, 67);"><i>just</i></span><span style="color:rgb(67, 67, 67);"> with rewards?</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The researchers were inspired by DeepSeek-R1-Zero, which used Reinforcement Learning (an approach which gives a computer program rewards for good behavior and penalties for bad). They wanted to see if they could do the same with <span style=""><i>multimodal</i></span> models (ones that understand both text and images). </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> They tried the simplest approach first: They took a bunch of math problems (10,000 of them), and used RL to train a basic MLLM. The MLLM got a reward if it: </p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Used the correct format:</b></span> The output had to be in a specific structure: "<think> [reasoning steps] </think> <answer> [answer] </answer>". Think of it like forcing the model to "show its work." </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Got the right answer:</b></span> The final answer had to match the correct solution to the problem. </p></li></ol></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> They called this model "Vision-R1-Zero." The result? It <span style=""><i>didn't</i></span> work very well. The model struggled to learn complex reasoning, and the reasoning it <span style=""><i>did</i></span> produce wasn't very long or sophisticated. Even worse, when they trained it for a long time, it started to get <span style=""><i>worse</i></span>, even though it was using longer reasoning steps. </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/514e153f-0d5d-4f38-b4ce-6d03d1419039/image.png?t=1742314272" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p>Data generation pipeline incorporating our Modality Bridging method</p></td></tr></table></td></tr><tr><td id="inner-workings-of-vision-r-1" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Inner Workings of Vision-R1 </span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Since the "RL-only" method failed, the researchers tried a different strategy. They combined two ideas: </p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Cold-Start Initialization:</b></span> First, they would "pre-train" the MLLM using a special dataset of multimodal problems that <span style=""><i>already included</i></span> the reasoning steps (the "Chain-of-Thought" or CoT). This is like giving the model a textbook with worked examples before asking it to solve problems on its own. This pre-trained model is called "Vision-R1-CI." </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Reinforcement Learning (again):</b></span> <span style=""><i>After</i></span> the cold-start, they would use RL to fine-tune the model. This time, the RL is used to help the model learn the <span style=""><i>correct</i></span> reasoning process, not to teach it reasoning from scratch. </p></li></ol></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The final model, after both steps, is called "Vision-R1." The key to the cold-start is having a good dataset with examples of human-like reasoning. Existing datasets were often too simple and didn't show the kind of back-and-forth thinking that humans do (like questioning assumptions or correcting mistakes). </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/081a34b0-f8a5-4e6e-bafe-19a4fba61e1f/image.png?t=1742314309" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p>Comparison between the CoT processes generated by descriptions with and without “Pseudo-CoT”.</p></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Researchers wanted to use the reasoning abilities of DeepSeek-R1 (the text-only model that <span style=""><i>was</i></span> good at reasoning) to help create this dataset. But DeepSeek-R1 can't understand images! </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Here's their clever solution, called "Modality Bridging": </p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Pseudo-CoT:</b></span> They took an image, a question, and the correct answer, and fed them to a <span style=""><i>different</i></span> MLLM. They asked this MLLM to generate a "Pseudo-CoT" – a description of the image <span style=""><i>and</i></span> some initial reasoning steps. This is like asking a student to explain the problem and take a first stab at solving it. </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Detailed Description:</b></span> They then took the original image and question, <span style=""><i>plus</i></span> the "Pseudo-CoT," and fed them <span style=""><i>back</i></span> into the MLLM. This time, they asked for a very detailed description of the image, <span style=""><i>incorporating</i></span> the information from the Pseudo-CoT. This is like asking a student to refine their explanation after getting some hints. The Pseudo-CoT helps the MLLM focus on the <span style=""><i>important</i></span> visual details. </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>DeepSeek-R1's Magic:</b></span> Now they had a detailed <span style=""><i>textual</i></span> description that included all the important visual information. They fed this description to DeepSeek-R1 (the text-only reasoning expert). DeepSeek-R1 could then generate high-quality, human-like reasoning steps (the CoT). </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Clean Up:</b></span> They filtered out any reasoning that led to the wrong answer, and did some minor cleaning to make the text more consistent. </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Putting it Together:</b></span> Finally, they combined the original image with the high-quality CoT generated by DeepSeek-R1. This created the "Vision-R1-cold" dataset – a collection of multimodal problems with excellent reasoning examples. </p></li></ol></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> After the cold-start initialization (using the Vision-R1-cold dataset), the model ("Vision-R1-CI") had learned <span style=""><i>how</i></span> to reason in a complex way. But it had a new problem: "Overthinking." </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/050a075d-e41c-43c0-8d92-dda7e746b2a1/image.png?t=1742314356" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p>GRPO with the proposed PTST strategy. </p></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The model would sometimes spend <span style=""><i>too much</i></span> time thinking, even when a shorter reasoning process would have been correct. The correct reasoning was often shorter and simpler. This made it hard for the RL training (in the next step) to work properly, because the model was getting "lost" in long, incorrect reasoning chains. </p></td></tr><tr><td id="teaching-the-model-to-think-efficie" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Teaching the Model to Think Efficiently via Progressive Thinking Suppression Training </span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> To solve the "overthinking" problem, they came up with "Progressive Thinking Suppression Training" (PTST). The idea is simple but powerful: </p></td></tr><tr><td style="padding-bottom:12px;padding-left:50px;padding-right:40px;padding-top:12px;" class="ee"><div style="margin-left:0px;" class="edm_outlooklist"><ol start="1" style="list-style-type:decimal;margin:0px 0px;padding:0px 0px 0px 0px;"><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Start Short:</b></span> At the beginning of RL training, they <span style=""><i>forced</i></span> the model to use short reasoning processes. This prevented it from getting lost in long, incorrect chains. It's like making the model practice solving easy problems first. </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Gradually Lengthen:</b></span> As training went on, they slowly allowed the model to use longer and longer reasoning. This gave the model time to learn the <span style=""><i>correct</i></span> reasoning patterns before it was allowed to explore more complex possibilities. </p></li><li class="listItem ultext"><p style="line-height:24px;padding:0px;text-align:left;word-break:break-word;"><span style="font-weight:700;"><b>Hard Formatting Reward:</b></span> They used a strict reward system. The model <span style=""><i>only</i></span> got a reward if it used the correct format <span style=""><i>and</i></span> got the right answer. No partial credit! </p></li></ol></div></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> They used a specific RL algorithm called "Group Relative Policy Optimization" (GRPO), but the key idea is the <span style=""><i>progressive</i></span> increase in allowed reasoning length. </p></td></tr><tr><td align="center" valign="top" style="padding: 20px 28px 20px; " class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 auto;"><tr><td align="center" valign="top" style="width:600px;"><img src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/f40c8636-7224-4c87-a594-5fbc043525d9/image.png?t=1742314414" alt="" height="auto" width="600" style="display:block;width:100%;" border="0"/></td></tr><tr><td align="center" valign="top" class="t" style="width:600px;"><p>The output examples of Vision-R1-7B on MathVerse benchmark. </p></td></tr></table></td></tr><tr><td id="real-world-implications-of-vision-r" class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:normal;padding:0px 28px;text-align:left;"><h3 style="color:#2A2A2A;font-weight:normal;"><span style="color:rgb(67, 67, 67);">Real-World Implications of Vision-R1</span></h3></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The resulting Vision-R1-7B model had <span style="font-weight:700;"><b>impressive mathematical reasoning capabilities</b></span>. On the MathVista benchmark, Vision-R1-7B nearly matches OpenAI's O1, a leading reasoning model, despite being <span style="font-weight:700;"><b>significantly smaller</b></span> (7B parameters vs. O1's unspecified, but likely much larger, size). It shows substantial gains (+10% accuracy) over its base model (Qwen-2.5-VL-7B) on challenging sub-tasks, indicating a strong grasp of complex, "human-like" reasoning. Vision-R1-7B also achieves top or <span style="font-weight:700;"><b>near-top performance</b></span> on other difficult math benchmarks (MathVerse and MM-Math). </p></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> The <span style=""><i>Vision-R1-cold</i></span> dataset, used for the model's initial training, is shown to be of high quality, containing significantly more instances of human-like cognitive processes (questioning, reflection, etc.) than previous datasets like Mulberry and LLaVA-CoT. This is demonstrated quantitatively and confirmed by the fact that a model trained on <span style=""><i>Vision-R1-cold</i></span> (Vision-R1-LlamaV-CI-11B) <span style="font-weight:700;"><b>outperforms</b></span> models trained on other datasets, achieving state-of-the-art results across various benchmarks. </p></td></tr><tr><td align="center" valign="top" style="padding-bottom:14px;padding-left:28px;padding-right:28px;padding-top:14px;text-align:center;width:100%;word-break:break-word;" class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" align="center" style=""><tr><td align="center" valign="middle" height="42" style="height:42px;background-color:#2C81E5;border-radius:10px;border:solid 0px #DFD150;color:#FFFFFF;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.DUiN96-Eq7pUHzwEhy5j28yjf9KIXZdsXoh1WlHvvKmD1wA42KfFESkjpaMwyw0uaNww5GWlDmhnvv8ttdBZfssLLsDr8giyfRCVcNPAB5m6BafgoJiOX62klZZY-_V1/4ex/Or5yX8RKQt6FNRtIx1tHyg/h15/h001.i7KYaZPU4AdgtjYmlDXow-HmYPIEEkVonw236TAqW8U" target="_blank" rel="noopener noreferrer nofollow" style="color:#FFFFFF;display:block;font-size:16px;padding:0px 14px;text-decoration:none;"> Read Full Paper </a></td></tr></table></td></tr><tr><td align="center" valign="top"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td align="center" valign="top" style="font-size:0px;line-height:0px;padding:30px 0px;" class="dd"><table class="j" role="none" width="50%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td> </td></tr></table></td></tr><tr><td class="dd" align="left" valign="top" style="color:#2A2A2A;font-weight:Bold;padding:0px 28px;text-align:left;"><h2 style="color:#2A2A2A;font-weight:Bold;">Share The AI Timeline</h2></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> You currently have <strong>0</strong> referrals. </p></td></tr><tr><td align="left" valign="top" style="padding: 20px 28px 20px; display:none;width:0px;max-height:0px;overflow:hidden;mso-hide:all;height:0;font-size:0;max-height:0;line-height:0;margin:0 auto;" class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" style="margin:0 auto 0 0;"><tr><td align="center" valign="top" style="width:300px;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.c6q0w4g5sodbtO4I1B_pxWc4htTObwdorovK0nFHVH-4pUdVE0ELYH5DsNemk732SjNwhPNJ25r0O8B5vYifsGNUqyW5TiZkyMsF1yreu0byy2KW36J1wDdpoLuXg2TU1F1OW8OHoHaU4-ZmrZpPU4RN-crQCEimD190CSn9fPuxpIRojBJyu1VfV5KtQD3QMVdSg2JrjEj5-xm4r4E12Whf08itqPCb9Q5W0X4rt3ubYkqCmWnLeZpmb3_RZcbIk0UE5wZnFLCQJHLFs0qZ0OGpXp89o1HU4mWIBur5Or4tQGm5M_Y8m5PvTEfYfxLRyrcRv7GyVs5oLtFfiySZ2SqtZypLA-h50h61p0uPiA7iA_PiMqlVLtM-87XL33VZi05_O3UTpWE_0nAzFRJ4TW1ayz3_vn4Zlp9IERdbnnAd_1kPLD4lAQcR5PRXgtpCs1MOfydRNnZx3s9QoAXQakWmk3LRRhtb932-I3tqFTwRipRdz7_0DdlfyXC3JNcsGnHqxmVe4A9C2Ki8rcr_rpV3iaNemCbCJ_Dz1gX7P1OjkC1_sGU6NHGw-r4tciPQH_3SPcBgZVAt0DdCWjPlEYDDao2TBCCPHGK6JSeaIpTI7TfOGaOwKPBlPbUwR61M/4ex/Or5yX8RKQt6FNRtIx1tHyg/h16/h001.srP6jVMYT9351sqOtH5fHucxjutabnvEL-bA5uV_B-4" rel="noopener noreferrer nofollow" style="text-decoration:none;" target="_blank"><img src="" alt="" height="auto" width="300" style="display:block;width:100%;" border="0"/></a></td></tr></table></td></tr><tr><td align="left" valign="top" style="padding-bottom:14px;padding-left:28px;padding-right:28px;padding-top:14px;text-align:left;width:100%;word-break:break-word;" class="dd"><table role="none" border="0" cellspacing="0" cellpadding="0" align="left" style=""><tr><td align="center" valign="middle" height="42" style="height:42px;background-color:#2C81E5;border-radius:10px;border:solid 0px #DFD150;color:#FFFFFF;font-family:'Open Sans','Segoe UI','Apple SD Gothic Neo','Lucida Grande','Lucida Sans Unicode',sans-serif;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.c6q0w4g5sodbtO4I1B_pxWc4htTObwdorovK0nFHVH-4pUdVE0ELYH5DsNemk732SjNwhPNJ25r0O8B5vYifsGNUqyW5TiZkyMsF1yreu0byy2KW36J1wDdpoLuXg2TU1F1OW8OHoHaU4-ZmrZpPU4RN-crQCEimD190CSn9fPuxpIRojBJyu1VfV5KtQD3QMVdSg2JrjEj5-xm4r4E12Whf08itqPCb9Q5W0X4rt3ubYkqCmWnLeZpmb3_RZcbIk0UE5wZnFLCQJHLFs0qZ0OGpXp89o1HU4mWIBur5Or4tQGm5M_Y8m5PvTEfYfxLRyrcRv7GyVs5oLtFfiySZ2SqtZypLA-h50h61p0uPiA7iA_PiMqlVLtM-87XL33VZi05_O3UTpWE_0nAzFRJ4TW1ayz3_vn4Zlp9IERdbnnAd_1kPLD4lAQcR5PRXgtpCs1MOfydRNnZx3s9QoAXQakWmk3LRRhtb932-I3tqFTwRipRdz7_0DdlfyXC3JNcsGnHqxmVe4A9C2Ki8rcr_rpV3iaNemCbCJ_Dz1gX7P1OjkC1_sGU6NHGw-r4tciPQH_3SPcBgZVAt0DdCWjPlEYDDao2TBCCPHGK6JSeaIpTI7TfOGaOwKPBlPbUwR61M/4ex/Or5yX8RKQt6FNRtIx1tHyg/h17/h001.SD4ArH0A9rCBIX1JeFIBJ2Cao-0IfB3pJFRKH8hvybo" target="_blank" rel="noopener noreferrer nofollow" style="color:#FFFFFF;display:block;font-size:16px;padding:0px 14px;text-decoration:none;"> Click to Share </a></td></tr></table></td></tr><tr><td class="dd" align="left" style="padding:0px 28px;text-align:left;word-break:break-word;"><p style="line-height:24px;"> Or copy and paste this link to others: <a class="link" href="https://mail.bycloud.ai/subscribe?ref=6SqUHb8KiF&_bhlid=7fecfad9eb7fd8bcdb529e945e11346b5897acdc" target="_blank" rel="noopener noreferrer nofollow" clicktracking="off"><span>https://mail.bycloud.ai/subscribe?ref=6SqUHb8KiF</span></a></p></td></tr><tr><td align="center" valign="top" style="font-size:0px;line-height:0px;padding:30px 0px;" class="dd"><table class="j" role="none" width="50%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td> </td></tr></table></td></tr></table></td></tr></table></td></tr><tr><td class="b" align="center" valign="top" bgcolor="#2a2a2a" style="padding:0px;border-bottom-left-radius:0px;border-bottom-right-radius:0px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td align="center" valign="top" bgcolor="#73ddff" style="padding:12px"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td><span style="padding-left:1px;"></span></td><td align="center" valign="middle" width="75" style="width:75px;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.1muhFWIqieRYpaJ-FbWSCQqcWoV4NNHHr5SkP9THApWuHAAlWLQxI3Q_IqFmt_DcyAxeC8jDApCnHmMSBGpBb5sgtimvBYgxRX-Rp7s0F3LjCHoSwdhr83OBqRFhJ1y_/4ex/Or5yX8RKQt6FNRtIx1tHyg/h18/h001.t-z68BYfrt5Mjsz3tEkwpfQ04h8y2tM2vMmhElbsMSk" style="text-decoration:none;"><img width="22" alt="tw" border="0" style="display:block;max-width:22px;" src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/static_assets/x_dark.png"/></a></td><td align="center" valign="middle" width="75" style="width:75px;"><a href="https://elink4f7.mail.bycloud.ai/ss/c/u001.amatuKKICSickUKplYJXmBoQnQ9VXnB2zTxBG4HeHBgjMqVxpoXRdj01cjwyoVlHgiebEOgBvwHtevoVpsSvpn3Q1di2ml6sb3cBM-X6IStQbj_zQSVGWJ8AAmPw2en2/4ex/Or5yX8RKQt6FNRtIx1tHyg/h19/h001.7v1y9F6u3R2mFaUS9oTOUcwLeknSYUxXqbxJ9E9D0mk" style="text-decoration:none;"><img width="22" alt="yt" border="0" style="display:block;max-width:22px;" src="https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/static_assets/youtube_dark.png"/></a></td><td><span style="padding-left:1px;"></span></td></tr></table></td></tr><tr><td height="10" style="line-height:1px;font-size:1px;height:10px;"> </td></tr><tr><td class="w" align="center" valign="top" style="padding:15px;"><table role="none" width="100%" border="0" cellspacing="0" cellpadding="0" align="center"><tr><td align="center" valign="top"><p style="font-family:'Verdana',Geneva,sans-serif;color:#FFFFFF!important;"> Update your email preferences or unsubscribe <a class="link" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.c6q0w4g5sodbtO4I1B_pxWc4htTObwdorovK0nFHVH-4pUdVE0ELYH5DsNemk732SjNwhPNJ25r0O8B5vYifsBhEpz-DJgyVFmavJPa0OyKRRnvw4o7XGyvIv7PRofnm2A4qvJhTQYaCxd3gdeVLoXFgD10JJy9xaK6vys_YFlbgPDH7ZRVVrrw24xcQiXaOFfuXp_BYMqIjqYomW4bn0MO1gul9b88ZLdCC_aARVizcnZZnEcy2gk9cMQ1UX_iR4FXrlcRWBiBihcwcfiduSUGapeZQreF7Ote-vah2IQYw5pDG5-fthhT-bLeP5HewiJDrRzNCHzdAzGKiFPUqZHmgF56XJuC_l8WkCA4K1poXWHN26aN0JmRh7tnJAumRDDyn26R7hB-FOu9Uyx276bUZSSOuamriWpem6MWm2d7UGIFw2GagJP4cEJkjmJkVNXu4w87ReL0HcCELrDrIlx8kKOwmp2_ya5KyK8INGL7pIpP8EJlomGM_76lVD7KoiuNyujHn_Yrp69aBf7192VX5L_0YsrmjLCau0zAnZXRJLhPHlC58dbsD0_IOHQRk9JcRoz79gLvO_vOMzLguUGWVbGEYc1SkeBmVxJt171jwYJgI4X-iYkwkL8dTDzGF5W7e1dRVaV293AN7rAbK5GUKYiSPi-j4NuLgPjUwGwr7Sa7Diw7nkISsuJdFBXp-Qy6GiwZR_2EQp3Kc-AjIcv80QLnUjtZtEJn3ZDu0EZnRnE-lWStTEJ4EwqSeI54kYQjJR_vB3UkeCZMbaucGJLHlSLuxgwDjSExESsM8dBFWj5Sj0ZwCFROywAnow5RnqqRIdisfdmYJ-3oa55A_qQ/4ex/Or5yX8RKQt6FNRtIx1tHyg/h20/h001.S8asyIU3tSGKg7Eg15pGApp3ErABdqI83_esMzH-B-E" style="text-decoration:underline;text-decoration-color:#FFFFFF!important;color:#FFFFFF!important;"> here</a></p><p class="copyright" style="font-family:'Verdana',Geneva,sans-serif;color:#FFFFFF!important;"> © 2025 bycloudai </p><p style="font-family:'Verdana',Geneva,sans-serif;color:#FFFFFF!important;"> 228 Park Ave S, #29976, New York, New York 10003, United States </p></td></tr><tr style="display: table-row !important;"><td align="center" valign="top" style="padding-top:20px;" style="display:table-cell !important;"><table role="none" border="0" cellspacing="0" cellpadding="0" align="center" style="display:table !important;"><tr style="display:table-row !important;"><td class="u" align="center" valign="middle" height="32" style="height:32px;display:table-cell !important; max-height: 32px !important;margin:0px !important; background-color: #ffffff !important;"><a style="line-height:32px !important;text-decoration:none;display:block !important;" href="https://elink4f7.mail.bycloud.ai/ss/c/u001.DUiN96-Eq7pUHzwEhy5j28olDWFpV5DDKfdk_OdOKOheIr4iqadjBqUEyfb2KARZS7UKyahx7uSPgDC0DaGsvgr9Dg7iuHY79ueOHYDu6-lJulk-zPihbzNg14RtUyfsZd60MATQ4bAWBHs9ubwC1JiA7hwrimYurbSVaZ4SvySNHy_R8iNfsiOtzo7XZ2gY4wtLmFwtQ-uqi-whTG0MajaXfm8LqnqictErygKWc1U83QEZ3VLA5JUwJ8hF-Im8/4ex/Or5yX8RKQt6FNRtIx1tHyg/h21/h001.MlCxgqYAsl2hsyIfzWs6XdajoB4ziqzTPk_67UCktP0"><img src="https://media.beehiiv.com/output-onlinepngtools.png" width="16" alt="beehiiv logo" style="display:inline-block !important;max-width:16px !important; vertical-align:-3px !important;width: 16px !important;" border="0"/><span style="padding-left:11px !important;display: inline-block !important;">Powered by beehiiv</span></a></td></tr></table></td></tr><tr><td align="left" valign="top" height="2" style="height:2px;"><a href='https://elink4f7.mail.bycloud.ai/ss/c/u001.CxDkkVpJsBdVoe83c_tBWsHIaP4XNp0WgUYqLvHcKk_3uqk_KIkz4ddLinhFbud6JuxLFdSUhYnR7b1NSsmbtzXNGNblnEEMKUtkCAjkn8Y/4ex/Or5yX8RKQt6FNRtIx1tHyg/h22/h001.VZNlsqIstn8TVIapD5zCBFXSLyhKZtOpAjJU6cDEkQU' style="color: #2a2a2a !important; cursor: default; font-size: 1px; text-decoration: none;"> Terms of Service </a></td></tr></table></td></tr></table></td></tr></table></td></tr></table></td></tr></table></td></tr></table></div></body></html>