Nikita Balagansky

26 papers A* 6Journal 18Unranked 2

Year	Rank	Type	Title / Venue / Authors
2026	J	`jnl`	Next Embedding Prediction Makes World Models Stronger. CoRR George Bredis, Nikita Balagansky, Daniil Gavrilov, Ruslan Rakhimov
2025	A*	`conf`	Analyze Feature Flow to Enhance Interpretation and Steering in Language Models. ICML Daniil Laptev, Nikita Balagansky, Yaroslav Aksenov, Daniil Gavrilov
2025	J	`jnl`	Analyze Feature Flow to Enhance Interpretation and Steering in Language Models. CoRR Daniil Laptev, Nikita Balagansky, Yaroslav Aksenov, Daniil Gavrilov
2025	J	`jnl`	Guided Star-Shaped Masked Diffusion. CoRR Viacheslav Meshchaninov, Egor Shibaev, Artem Makoian, Ivan Klimov, Danil Sheshenya, Andrey Malinin, Nikita Balagansky, Daniil Gavrilov, Aibek Alanov, Dmitry P. Vetrov
2025	A*	`conf`	Learn Your Reference Model for Real Good Alignment. ICLR Alexey Gorbatovski, Boris Shaposhnikov, Alexey Malakhov, Nikita Surnachev, Yaroslav Aksenov, Ian Maksimov, Nikita Balagansky, Daniil Gavrilov
2025	A*	`conf`	Mechanistic Permutability: Match Features Across Layers. ICLR Nikita Balagansky, Ian Maksimov, Daniil Gavrilov
2025	J	`jnl`	Small Vectors, Big Effects: A Mechanistic Study of RL-Induced Reasoning via Steering Vectors. CoRR Viacheslav Sinii, Nikita Balagansky, Gleb Gerasimov, Daniil Laptev, Yaroslav Aksenov, Vadim Kurochkin, Alexey Gorbatovski, Boris Shaposhnikov, Daniil Gavrilov
2025	A*	`conf`	Steering LLM Reasoning Through Bias-Only Adaptation. EMNLP Viacheslav Sinii, Alexey Gorbatovski, Artem Cherepanov, Boris Shaposhnikov, Nikita Balagansky, Daniil Gavrilov
2025	J	`jnl`	Steering LLM Reasoning Through Bias-Only Adaptation. CoRR Viacheslav Sinii, Alexey Gorbatovski, Artem Cherepanov, Boris Shaposhnikov, Nikita Balagansky, Daniil Gavrilov
2025	J	`jnl`	Teach Old SAEs New Domain Tricks with Boosting. CoRR Nikita Koriagin, Yaroslav Aksenov, Daniil Laptev, Gleb Gerasimov, Nikita Balagansky, Daniil Gavrilov
2025	A*	`conf`	Train One Sparse Autoencoder Across Multiple Sparsity Budgets to Preserve Interpretability and Accuracy. EMNLP Nikita Balagansky, Yaroslav Aksenov, Daniil Laptev, Vadim Kurochkin, Gleb Gerasimov, Nikita Koriagin, Daniil Gavrilov
2025	J	`jnl`	Train One Sparse Autoencoder Across Multiple Sparsity Budgets to Preserve Interpretability and Accuracy. CoRR Nikita Balagansky, Yaroslav Aksenov, Daniil Laptev, Vadim Kurochkin, Gleb Gerasimov, Nikita Koryagin, Daniil Gavrilov
2025	J	`jnl`	Train Sparse Autoencoders Efficiently by Utilizing Features Correlation. CoRR Vadim Kurochkin, Yaroslav Aksenov, Daniil Laptev, Daniil Gavrilov, Nikita Balagansky
2025	J	`jnl`	VARAN: Variational Inference for Self-Supervised Speech Models Fine-Tuning on Downstream Tasks. CoRR Daria Diatlova, Nikita Balagansky, Alexander Varlamov, Egor Spirin
2025	J	`jnl`	You Do Not Fully Utilize Transformer's Representation Capacity. CoRR Gleb Gerasimov, Yaroslav Aksenov, Nikita Balagansky, Viacheslav Sinii, Daniil Gavrilov
2024	J	`jnl`	Learn Your Reference Model for Real Good Alignment. CoRR Alexey Gorbatovski, Boris Shaposhnikov, Alexey Malakhov, Nikita Surnachev, Yaroslav Aksenov, Ian Maksimov, Nikita Balagansky, Daniil Gavrilov
2024	—	`conf`	Linear Transformers with Learnable Kernel Functions are Better In-Context Models. ACL (1) Yaroslav Aksenov, Nikita Balagansky, Sofia Maria Lo Cicero Vaina, Boris Shaposhnikov, Alexey Gorbatovski, Daniil Gavrilov
2024	J	`jnl`	Linear Transformers with Learnable Kernel Functions are Better In-Context Models. CoRR Yaroslav Aksenov, Nikita Balagansky, Sofia Maria Lo Cicero Vaina, Boris Shaposhnikov, Alexey Gorbatovski, Daniil Gavrilov
2024	J	`jnl`	Mechanistic Permutability: Match Features Across Layers. CoRR Nikita Balagansky, Ian Maksimov, Daniil Gavrilov
2023	J	`jnl`	Ahead-of-Time P-Tuning. CoRR Daniil Gavrilov, Nikita Balagansky
2023	J	`jnl`	Democratized Diffusion Language Model. CoRR Nikita Balagansky, Daniil Gavrilov
2022	J	`jnl`	Classifiers are Better Experts for Controllable Text Generation. CoRR Askhat Sitdikov, Nikita Balagansky, Daniil Gavrilov, Alexander Markov
2022	J	`jnl`	Linear Interpolation In Parameter Space is Good Enough for Fine-Tuned Language Models. CoRR Mark Rofin, Nikita Balagansky, Daniil Gavrilov
2022	A*	`conf`	PALBERT: Teaching ALBERT to Ponder. NeurIPS Nikita Balagansky, Daniil Gavrilov
2022	—	`conf`	PALBERT: Teaching ALBERT to Ponder. RepL4NLP@ACL Daniil Gavrilov, Nikita Balagansky
2022	J	`jnl`	PALBERT: Teaching ALBERT to Ponder. CoRR Nikita Balagansky, Daniil Gavrilov