Shin'ya Yamaguchi
@syamaguchi_en
Followers
162
Following
499
Media
35
Statuses
235
Researcher at NTT, PhD in Informatics. Research interests: generative models, dataset synthesis, and multi-modal LLMs.
Joined February 2023
#AcademicTwitter Shout out to all who are #PhDone in September. You are all awesome! 😀💪@iAvimanyu
@hsynsaltan @Au_Erithacus @QuantumYakar @Pedro18_Neto @ZuzanaMachacova @musicaesacrae @Supriya_Chem @syamaguchi_en @alican_kusoglu
0
3
9
#PhDone ! I've finally completed my PhD in Informatics! A huge thank to my advisor, Hisashi Kashima, and to my collaborators at NTT and Kyoto University.
11
10
224
We’ve received A LOT OF submissions this year 🤯🤯 and are excited to see so much interest! To ensure high-quality review, we are looking for more dedicated reviewers. If you'd like to help, please sign up here
docs.google.com
Please use this form to volunteer or nominate others to be ICLR '26 Reviewers. Please note that the final decision to invite Reviewers will be at the discretion of the program chairs.
12
72
362
We apologize for the delay in releasing #WACV2026 Round 1 decisions. The Program Chairs expect to release them by the end of the day today.
1
3
40
1/3) I am biased, but I think this is going to be big! CoVAE: Consistency Training of Variational Autoencoders We unify consistency models with VAEs to obtain a powerful and elegant generative autoencoder! The brainchild of the brilliant @gisilvs (who is looking for jobs!)
7
36
290
EurIPS is coming! 📣 Mark your calendar for Dec. 2-7, 2025 in Copenhagen 📅 EurIPS is a community-organized conference where you can present accepted NeurIPS 2025 papers, endorsed by @NeurIPSConf and #NordicAIR and is co-developed by @ELLISforEurope
https://t.co/RSAvf9lcZm
1
27
76
Every ML Engineer’s dream loss curve: “Kimi K2 was pre-trained on 15.5T tokens using MuonClip with zero training spike, demonstrating MuonClip as a robust solution for stable, large-scale LLM training.” https://t.co/IUbXZ6Hu7M
🚀 Hello, Kimi K2! Open-Source Agentic Model! 🔹 1T total / 32B active MoE model 🔹 SOTA on SWE Bench Verified, Tau2 & AceBench among open models 🔹Strong in coding and agentic tasks 🐤 Multimodal & thought-mode not supported for now With Kimi K2, advanced agentic intelligence
21
205
2K
Tokenization is just a special case of "chunking" - building low-level data into high-level abstractions - which is in turn fundamental to intelligence. Our new architecture, which enables hierarchical *dynamic chunking*, is not only tokenizer-free, but simply scales better.
Tokenization has been the final barrier to truly end-to-end language models. We developed the H-Net: a hierarchical network that replaces tokenization with a dynamic chunking process directly inside the model, automatically discovering and operating over meaningful units of data
61
195
1K
Our new work on adaptive image tokenization: Image —> T tokens * variable T, based on image complexity * single forward pass both infers T and tokenizes to T tokens * approximates minimum description length encoding of the image
Compression is the heart of intelligence From Occam to Kolmogorov—shorter programs=smarter representations Meet KARL: Kolmogorov-Approximating Representation Learning. Given an image, token budget T & target quality 𝜖 —KARL finds the smallest t≤T to reconstruct it within 𝜖🧵
0
30
202
What concerns were addressed by rebuttal (both yours and other reviewers)? What remain? Finally, summarize why you recommend accept or reject.
2
1
3
Our paper "Influential Bandits: Pulling an Arm May Change the Environment" has been accepted to #TMLR 🎉 We proposed and analyzed the influential bandit problem, where an action affects the rewards of other actions. Paper📜： https://t.co/rBRTlb9gJ2
0
3
24
This is also an awesome work by Ryota Tanaka @rtanaka_lab , enabling visually document processing by RAG with related textual images! Come NOW to #363 at #CVPR2025 poster session!
0
1
7
This is happening NOW at #390!
🚀 Happy to share our @CVPR paper, which proposes a post-pre-training method for CLIP to mitigate modality gaps and improve zero-shot performance with just 5 minutes of additional training! We are looking forward to discussing with you at our poster session! #CVPR2025
0
0
2