Researcher — paper.dinq.me

2026 顶会论文

作者位次 #3/6NeurIPS 2026Poster· Social Aspects / Accountability, Transparency and Interpretability

Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?

Anton Korznikov · Andrey V. Galichin · Alexey Dontsov · Oleg Rogov · Ivan Oseledets · Elena Tutubalina

可解释性社会与人本OpenReview

作者位次 #7/10ACL 2026Poster· ACL Industry Track

Motivating Next-Gen Accelerators with Flexible N{:}M Activation Sparsity via Benchmarking Lightweight Post-Training Sparsification Approaches

Shirin Alanova · Kristina Kazistova · Ekaterina Galaeva · Alina Kostromina · Vladimir Smirnov · Redko Dmitry · Alexey Dontsov · Maxim Zhelnin · Evgeny Burnaev · Egor Shvetsov

摘要

The demand for efficient large language model inference has spurred interest in sparsification, yet current hardware support remains narrowly focused on 2:4 weight sparsity. In this work, we argue that activation sparsity despite being overlooked in hardware design offers a promising path for dyn...

效率与系统评测与基准强化学习与后训练OpenReview 项目页

作者位次 #2/7AAAI 2026Poster· Artificial Intelligence

I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders

Andrey V. Galichin · Alexey Dontsov · Polina Druzhinina · Anton Razzhigaev · Oleg Rogov · Elena Tutubalina · Ivan V. Oseledets

可解释性推理OpenReview

Alexey Dontsov

获取完整研究档案

联系方式

2026 顶会论文

Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?

Motivating Next-Gen Accelerators with Flexible N{:}M Activation Sparsity via Benchmarking Lightweight Post-Training Sparsification Approaches

I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders

常见合作者