Janos Kramar

Researcher, DeepMind

Joined

December 2016

Names

Janos Kramar (Preferred)

János Kramár

Emails

****@gmail.com (Confirmed)

****@polymtl.ca

****@google.com (Confirmed)

Personal Links

DBLP

Career & Education History

Researcher

DeepMind (deepmind.com)

2017 – Present

Researcher

Polytechnique Montreal (polymtl.ca)

2016 – 2016

PhD student

Harvard University (harvard.edu)

2010 – 2012

Advisors, Relations & Conflicts

No relations added

Expertise

No areas of expertise listed

Publications

Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
Senthooran Rajamanoharan, Tom Lieberum, Nicolas Sonnerat, Arthur Conmy, Vikrant Varma, Janos Kramar, Neel Nanda
- Submitted to ICLR 2025
- Readers: Everyone
Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2
Tom Lieberum, Senthooran Rajamanoharan, Arthur Conmy, Lewis Smith, Nicolas Sonnerat, Vikrant Varma, Janos Kramar, Anca Dragan, Rohin Shah, Neel Nanda
- BlackboxNLP 2024
- Readers: Everyone
Improving Sparse Decomposition of Language Model Activations with Gated Sparse Autoencoders
Senthooran Rajamanoharan, Arthur Conmy, Lewis Smith, Tom Lieberum, Vikrant Varma, Janos Kramar, Rohin Shah, Neel Nanda
- ICML 2024 MI Workshop Poster
- Readers: Everyone
Improving Sparse Decomposition of Language Model Activations with Gated Sparse Autoencoders
Senthooran Rajamanoharan, Arthur Conmy, Lewis Smith, Tom Lieberum, Vikrant Varma, Janos Kramar, Rohin Shah, Neel Nanda
- NeurIPS 2024 poster
- Readers: Everyone
On scalable oversight with weak LLMs judging strong LLMs
Zachary Kenton, Noah Yamamoto Siegel, Janos Kramar, Jonah Brown-Cohen, Samuel Albanie, Jannis Bulian, Rishabh Agarwal, David Lindner, Yunhao Tang, Noah Goodman, Rohin Shah
- NeurIPS 2024 poster
- Readers: Everyone
Explaining grokking through circuit efficiency
Vikrant Varma, Rohin Shah, Zachary Kenton, Janos Kramar, Ramana Kumar
- Submitted to ICLR 2024
- Readers: Everyone
Tracr: Compiled Transformers as a Laboratory for Interpretability
David Lindner, Janos Kramar, Sebastian Farquhar, Matthew Rahtz, Thomas McGrath, Vladimir Mikulik
- NeurIPS 2023 spotlight
- Readers: Everyone
Should I tear down this wall? Optimizing social metrics by evaluating novel actions.
János Kramár, Neil C. Rabinowitz, Tom Eccles, Andrea Tacchetti
- 2020 (modified: 09 Nov 2022)
- CoRR2020
- Readers: Everyone
Learning to Play No-Press Diplomacy with Best Response Policy Iteration
Thomas W. Anthony, Tom Eccles, Andrea Tacchetti, János Kramár, Ian M. Gemp, Thomas C. Hudson, Nicolas Porcel, Marc Lanctot, Julien Pérolat, Richard Everett, Satinder Singh, Thore Graepel, Yoram Bachrach
- 2020 (modified: 04 Nov 2022)
- NeurIPS 2020
- Readers: Everyone
Learning to Play No-Press Diplomacy with Best Response Policy Iteration
Thomas W. Anthony, Tom Eccles, Andrea Tacchetti, János Kramár, Ian M. Gemp, Thomas C. Hudson, Nicolas Porcel, Marc Lanctot, Julien Pérolat, Richard Everett, Satinder Singh, Thore Graepel, Yoram Bachrach
- 2020 (modified: 04 Nov 2022)
- CoRR 2020
- Readers: Everyone

View all 21 publications

Co-Authors

View all 91 co-authors