Apple is presenting new analysis on the annual convention on Worldwide Convention on Studying Representations (ICLR), which takes place in individual in Singapore from April 24 to twenty-eight. We’re proud to once more sponsor the convention, which brings collectively the scientific and industrial analysis communities in deep studying. Under is an summary of Apple’s participation at ICLR 2025.
Schedule
Cease by the Apple sales space (#C03) within the Singapore EXPO throughout exhibition hours:
- Thursday, April 24: 09:30 – 17:30
- Friday, April 25: 09:30 – 17:30
- Saturday, April 26: 09:30 – 17:30
All occasions listed in GMT +8 (Singapore time).
Thursday, April 24
- POSTER
- The best way to Confirm Any (Affordable) Distribution Property: Computationally Sound Argument Techniques for Distributions
- 10:00 – 12:30, #30266, Poster Session 1, Corridor 3 + Corridor 2B
- Tal Herman (Weizmann Institute of Science), Man Rothblum
- POSTER
- Step-by-Step Reasoning for Math Issues through Twisted Sequential Monte Carlo
- 10:00 – 12:30, #29208, Poster Session 1, Corridor 3 + Corridor 2B
- Shengyu Feng (CMU), Xiang Kong, Shuang Ma, Aonan Zhang, Dong Yin, Chong Wang, Ruoming Pang, Yiming Yang (CMU)
- POSTER
- SeedLM: Compressing LLM Weights via Seeds of a Pseudo-Random Generator
- 15:00 – 17:30, #28000, Poster Session 2, Corridor 3 + 2B
- Rasoul Shafipour, David (MIND) Harrison, Max Horton, Jeff Marker, Houman Bedayat, Sachin Mehta (Meta), Mohammad Rastegari (Meta), Mahyar Najibi, Saman Naderiparizi
Friday, April 25
- POSTER
- EC-DIT: Scaling Diffusion Transformers with Knowledgeable Selection Routing
- 10:00 – 12:30, #29721, Poster Session 3, Corridor 3 + Corridor 2B
- Haotian Solar (Georgia Institute of Know-how), Tao Lei, Bowen Zhang, Yanghao Li, Haoshuo Huang, Ruoming Pang, Bo Dai (Georgia Institute of Know-how), Nan Du
- POSTER
- Ferret-UI 2: Mastering Common Person Interface Understanding Throughout Platforms
- 10:00 – 12:30, #32099, Poster Session 3, Corridor 3 + Corridor 2B
- Zhangheng Li, Eager You, Haotian Zhang, Di Feng, Harsh Agrawal, Xiujun Li, Mohana Prasad Sathya Moorthy, Jeff Nichols, Yinfei Yang, Zhe Gan
- POSTER
- Massive-Scale Picture-Caption Information in Bettering Multimodal Basis Fashions
- 10:00 – 12:30, #29536, Poster Session 3, Corridor 3 + Corridor 2B
- Jeff Lai, Vasileios Saveris, Chen Chen, Hong-You Chen, Haotian Zhang, Bowen Zhang, Wenze Hu, Juan Lao Tebar, Zhe Gan, Peter Grasch, Meng Cao, Yinfei Yang
- POSTER
- MM1.5: Strategies, Evaluation & Insights from Multimodal LLM Nice-tuning
- 10:00 – 12:30, #30222, Poster Session 3, Corridor 3 + Corridor 2B
- Haotian Zhang, Mingfei Gao, Zhe Gan, Philipp Dufter, Nina Wenzel, Forrest Huang, Dhruti Shah, Xianzhi Du, Bowen Zhang, Yanghao Li, Sam Dodge, Eager You, Zhen Yang, Aleksei Timofeev, Mingze Xu, Hong-You Chen, Jean-Philippe Fauconnier Biard, Jeff Lai, Haoxuan You, Zirui Wang, Afshin Dehghan, Peter Grasch, Yinfei Yang
- POSTER
- MMEgo: In direction of Constructing Selfish Multimodal LLMs for Video QA
- 10:00 – 12:30, #30907, Poster Session 3, Corridor 3 + Corridor 2B
- Hanrong Ye (Hong Kong College of Science and Know-how (HKUST)), Haotian Zhang, Erik Daxberger, Lin Chen, Zongyu Lin (UCLA), Yanghao Li, Bowen Zhang, Haoxuan You, Jiasen Lu, Dan Xu (HKUST), Zhe Gan, Yinfei Yang
- POSTER
- Principle, Evaluation, and Greatest Practices for Sigmoid Self-Consideration
- 10:00 – 12:30, #29205, Poster Session 3, Corridor 3 + 2B
- Jason Ramapuram, Federico Danieli, Eeshan Gunesh Dhekane, Floris Weers, Dan Busbridge, Pierre Ablin, Tatiana Likhomanenko, Jagrit Digani, Zijin Gu, Amitis Shidani, Russ Webb
- POSTER
- TIS-DPO: Token-level Significance Sampling for Direct Choice Optimization
- 10:00 – 12:30, #28368, Poster Session 3, Corridor 3 + Corridor 2B
- Legislation Liu (Tsinghua College), Felix Bai, Zhiyun Lu, Yanchao Solar, Xiang Kong, Simon Wang, Jiulong Shan, Lijie Wen (Tsinghua College), Philip S. Yu (College of Illinois at Chicago), Meng Cao
- SOCIAL
- Girls in Machine Studying (WiML)
- 12:30 – 14:00, Convention GHJ
- Helen Zhou and Nandita Bhaskhar will symbolize Apple on the WiML social.
- POSTER
- DART: Denoising Autoregressive Transformer for Scalable Textual content-to-Picture Era
- 15:00 – 17:30, #29151, Poster Session 4, Corridor 3 + Corridor 2B
- Jiatao Gu, Shuangfei Zhai, Yuyang Wang, Qihang Zhang (The Chinese language College of Hong Kong), Yizhe Zhang, Dinghuai Zhang (Mila), Navdeep Jaitly, Josh Susskind
- POSTER
- Do LLMs Know Internally When They Comply with Directions?
- 15:00 – 17:30, #28257, Poster Session 4, Corridor 3 + Corridor 2B
- Juyeon Heo (College of Cambridge), Christina Heinze-Deml, Oussama Elachqar, Shirley Ren, Udhay Nallasamy, Andy Miller, Kwan Ho Ryan Chan (College of Pennsylvania), Jaya Narain
- SOCIAL
- LatinX in AI
- 17:00 – 18:30, Convention GHJ
- Alejandro Newell and Miguel Sarabia del Castillo will symbolize Apple on the LatinX in AI social.
Saturday, April 26
- POSTER
- A Formal Framework for Understanding Size Generalization in Transformers
- 10:00 – 12:30, #29490, Poster Session 5, Corridor 3 + Corridor 2B
- Xinting Huang (Saarland College), Andy Yang (College of Notre Dame), Yash Sarrof (Saarland College), Mark Rofin (Saarland College), Satwik Bhattamishra (College of Oxford), Andreas Krebs (College of Tübingen), Hattie Zhou (MILA), Preetum Nakkiran, Michael Hahn (Saarland College)
- POSTER
- Scaling Diffusion Language Fashions through Adaptation from Autoregressive Fashions
- 10:00 – 12:30, #28659, Poster Session 5, Corridor 3 + Corridor 2B
- Shansan Gong (HKU), Shivam Agarwal (UIUC), Yizhe Zhang, Lin Zheng (HKU), Jiacheng Ye (HKU), Mukai Li (HKU), Chenxin An (HKU), Hao Peng (UIUC), Lingpeng Kong (HKU)
- POSTER
- RelCon: Relative Contrastive Studying for a Movement Basis Mannequin for Wearable Information
- 15:00 – 17:30, #28603, Poster Session 6, Corridor 3 + Corridor 2B
- Max Xu (UIUC), Jaya Narain, Greg Darnell, Hyewon Jeong (MIT), Haraldur Hallgrimsson, Darren Forde, Richard Fineman, James M. Rehg (UIUC), Karthik Jayaraman Raghuram, Shirley Ren
- SOCIAL
- Queer in AI
- 17:00 – 18:30, Convention GHJ
- Azim Yusoff, Kevin Miao, and Nate True will symbolize Apple on the Queer in AI social.
Sunday, April 27
Sunday, April 27
Monday, April 28
Technical Demos
Go to Apple’s sales space at Singapore EXPO, Sales space C03, to see our technical demos throughout exhibition hours:
- DEMO
- FastVLM
- FastVLM is a household of mobile-friendly imaginative and prescient language fashions. These on-device fashions use a mixture of CNN and transformer encoding methods. Designed particularly for on-device functions like chatbots, captions, and picture finders. Collectively, they optimize the stability between accuracy and pace.
- DEMO
- Depth Professional
- Zero-shot monocular depth estimation from photographs without having to know something concerning the digicam throughout coaching. Depth Professional can generalize to all kinds of photographs together with in-the-wild web photographs, low-light, textual content, and motion-blurred photographs from a smartphone. It makes use of a query-based structure to supply state-of-the-art imaginative and prescient transformer modeling, and it really works with each RGB and depth at a number of scales. Outcomes present that Depth Professional has unmatched functionality in out-of-domain generalization and accuracy, and it really works with every kind of photographs. Absolute depth cues in every native area are offered.
Accepted Papers
- A Formal Framework for Understanding Size Generalization in Transformers
- Xinting Huang (Saarland College), Andy Yang (College of Notre Dame), Yash Sarrof (Saarland College), Mark Rofin (Saarland College), Satwik Bhattamishra (College of Oxford), Andreas Krebs (College of Tübingen), Hattie Zhou (MILA), Preetum Nakkiran, Michael Hahn (Saarland College)
- The best way to Confirm Any (Affordable) Distribution Property: Computationally Sound Argument Techniques for Distributions
- Tal Herman (Weizmann Institute of Science), Man Rothblum
- Massive-Scale Picture-Caption Information in Bettering Multimodal Basis Fashions
- Jeff Lai, Vasileios Saveris, Chen Chen, Hong-You Chen, Haotian Zhang, Bowen Zhang, Wenze Hu, Juan Lao Tebar, Zhe Gan, Peter Grasch, Meng Cao, Yinfei Yang
- MM1.5: Strategies, Evaluation & Insights from Multimodal LLM Nice-tuning
- Haotian Zhang, Mingfei Gao, Zhe Gan, Philipp Dufter, Nina Wenzel, Forrest Huang, Dhruti Shah, Xianzhi Du, Bowen Zhang, Yanghao Li, Sam Dodge, Eager You, Zhen Yang, Aleksei Timofeev, Mingze Xu, Hong-You Chen, Jean-Philippe Fauconnier Biard, Jeff Lai, Haoxuan You, Zirui Wang, Afshin Dehghan, Peter Grasch, Yinfei Yang
- MMEgo: In direction of Constructing Selfish Multimodal LLMs for Video QA
- Hanrong Ye (Hong Kong College of Science and Know-how (HKUST)), Haotian Zhang, Erik Daxberger, Lin Chen, Zongyu Lin (UCLA), Yanghao Li, Bowen Zhang, Haoxuan You, Jiasen Lu, Dan Xu (HKUST), Zhe Gan, Yinfei Yang
- Principle, Evaluation, and Greatest Practices for Sigmoid Self-Consideration
- Jason Ramapuram, Federico Danieli, Eeshan Gunesh Dhekane, Floris Weers, Dan Busbridge, Pierre Ablin, Tatiana Likhomanenko, Jagrit Digani, Zijin Gu, Amitis Shidani, Russ Webb
Workshop Accepted Papers
Acknowledgements
Alexander Toshev and Ronan Collobert are Senior Space Chairs.
Chen Huang, Chong Wang, Eugene Ndiaye, Harsh Agrawal, Pau Rodriguez, Preetum Nakkiran, Stephan Richter, Yizhe Zhang, and Zhe Gan are Space Chairs.
Arno Blaas is a Workshop Co-Organizer, and Nicholas Apostoloff and Niv Sivakumar are Workshop Reviewers for “I Cannot Consider It is Not Higher: Challenges in Deep Studying (ICBINB) 2025”.
Agni Kumar, Andrew Szot, Arno Blaas, Barry Theobald, Bhuwan Dhingra, Devon Helm, Fartash Faghri, Hadi Pour Ansari, Haoxuan You, Huangjie Zheng, Iman Mirzadeh, Juri Minxha, Kunal Talwar, Lin Chen, Louis Bethune, Luca Zappella, Maartje ter Hoeve, Max Horton, Michael Kirchhof, Nicholas Apostoloff, Pavan Kumar Anasosalu Vasu, Philipp Krähenbühl, Pierre Ablin, Rasoul Shafipour, Raviteja Vemulapalli, Rin Metcalf Susa, Rupchen (Esther) Zhao, Santhosh Kumar Ramakrishnan, Vimal Thilak, Xavier Suau Cuadros, Xiaoming Zhao, and Zakhar Shumaylov are Reviewers.
