LAB LOGA page on LabLog, a review site (not the lab's official page)
Sato Lab
Machine-translated · View original
The lab uncovers memory, thinking, reasoning, and bias in machine learning from a theoretical foundation.
AboutAbout the lab
Understanding "intelligence" deeply through theory
Sato Lab studies "intelligence" in machine learning from the perspectives of memory, thinking, reasoning, and bias.
The lab aims to combine empirical research with the development of theory for solving real-world problems.
It values clearly defining the problems that society needs to solve and solving them through effective abstraction.
It places more emphasis on discovering new problems worth solving than on solution methods that rely on advanced mathematics.
Key terms
- Internal representation
- The way a trained model represents data such as images or text as sequences of numbers inside the computer
- Chain of Thought
- A method in which a language model is made to write out its intermediate reasoning step by step, not just the answer, in order to reason
- Inductive bias
- The built-in tendency of a model regarding which kinds of answers it is likely to choose when learning from limited data
NewsLab news
Paper accepted to ICLR 2026, an international conference on machine learning
Paper accepted to NeurIPS 2025, an international conference on machine learning
Paper accepted to COLM 2025, an international conference on language models
Paper accepted to ACL 2025, an international conference on natural language processing
Two papers accepted to ICML 2025, an international conference on machine learning
ResearchResearch themes
The lab's goal is to build a science of "memory, thinking, reasoning, and bias" for "intelligence" in machine learning. Through a deep understanding of these, it aims for the "emergence" of intelligence that lies beyond them.
*Illustrative imageUnderstanding How Memory Works
The lab investigates how deep models such as language models reorganize and store information inside the computer. It aims to clarify what makes a good internal representation and how to train models to acquire one.
*Illustrative imageExploring Reasoning and Thinking
Using approaches such as Chain of Thought, inference-time computation, and tuning with reinforcement learning, the lab analyzes the mechanisms that allow models to reason logically about complex problems and think flexibly. It aims to make the decision-making process of LLMs easier to see and to build highly accurate reasoning models.
*Illustrative imageAnalyzing Bias
Focusing on the inductive biases that arise during training, the lab analyzes what kinds of biases models have. It aims to build fair and trustworthy AI systems.
MembersFaculty
The members of Sato Lab consist of faculty, project researchers, and students in the doctoral, master's, and undergraduate programs.As of 2026
- ProfessorIssei SatoPublications ↗
- Assistant ProfessorYoshiaki KawasePublications ↗
VoicesStudent voices
Compiled from answers by students who were in the lab. Ratings and reviews reflect individual students' experiences, not the lab's official views.
Students who take part in an interview may receive a thank-you gift worth ¥1,000, regardless of what they say.
On a 4-point scale, ● marks where students' answers fall
Who is this lab for?
What's great about this lab?
What did you learn after joining?
What's the vibe, and what is the professor like?
Where do graduates go?
RelatedLabs similar to Sato Lab
Yanaka LabFaculty of Science, Department of Information ScienceComputational linguistics, natural language processing, and …
Miyao LabFaculty of Science, Department of Information ScienceResearch in natural language processing and computational li…
Ma LabFaculty of Science, Department of Information ScienceEngineering and quality assurance of highly reliable AI syst…
Yamasaki LabFaculty of Science, Department of Information ScienceQuantum information theory, quantum computation, quantum mac…The information on this page is compiled by the LabLog team from the lab's public information and interviews. For corrections or removal requests, please contact LabLog, not the lab.
Report an errorAbout the images
The top and research theme images are illustrative and were not taken at the lab. They are cropped from Wikimedia Commons images. Cropped CC BY-SA images are provided under the same CC BY-SA license as the original (version shown on each line).
- Top image: “Surface with path of greatest slope” LithiumFlash (see note below) / CC BY-SA 4.0 (cropped)
- Understanding How Memory Works: “T-SNE Embedding of MNIST” Kyle McDonald / CC BY 2.0 (cropped)
- Exploring Reasoning and Thinking: “Tic-tac-toe-game-tree” The original uploader was Gdr at English Wikipedia. / CC BY-SA 3.0 (cropped)
- Analyzing Bias: “Overfitting regression” GongonR / CC BY-SA 4.0 (cropped)
- Related lab (谷中研究室): “Bookshelves and interior of the library of the Academy for Judiciary” Szeronine / CC BY 4.0 (cropped)
- Related lab (宮尾研究室): “Rosetta Stone” Hans Hillewaert / CC BY-SA 4.0 (cropped)
- Related lab (馬研究室): “Unmarked Nissan Rogue - Possible Test Autonomous Vehicle - San Francisco, May 2025 04” 9yz / CC BY 4.0 (cropped)
- Related lab (山崎研究室): “Quantum Computer with Dispersive Background 002” OJB Quantum / CC BY 4.0 (cropped)
About LabLog
LabLog is a lab review app for students only. Student members at each university interview students in labs and publish what they say. Student voices can only be read by students who sign up with a university email address. Operated by bestiee Inc.
Learn more about LabLog ↗Search engines and AI may reference the basic lab info and rating summaries on this page and the description of LabLog. Student voices are for student members only. Automated collection by programs (scraping, or input to or training of AI) is prohibited under Article 8 of the Terms of Service. If you need data for research or analysis, please contact us.