
Basic Research of AI: Optimization in ML, Federated Learning, GenAI, LLM, Maths for AI, AI for Science Head: DrSc Aleksandr Beznosikov [email protected]
Joined September 2025
- Tweets78
- Following16
- Followers112
- Likes17
🌴 Mission: look under mHC’s hood
Multiple streams, but the model favors one. Disabling mixing in almost all layers barely changed perplexity.
We investigate stream collapse in our CIKM 2026 paper:
doi.org/10.1145/3799682.3840…
3
28
🔥 BRAIn Lab introduces SimplexLoRA: a new method for more efficient LLM fine-tuning.
Standard LoRA allocates resources evenly across layers. SimplexLoRA focuses them on task-relevant layers — improving quality under the same compute budget.
arxiv.org/abs/2610.37206
#EMNLP2026
Hi Budapest! BRAIn Lab is at @emnlpmeeting
We are here with 4 accepted works on LLM compression, quantization, refusal directions and LoRA.
If you are here - stop by our poster!
8
195
Replying to @tinkerapi
Nice to see this work! Related Muon-to-LoRA methods include our lab’s LoRA-TSD, Riemannion, and LoRA-Muon. LoRA-TSD uses Muon-style spectral updates in LoRA’s tangent space, with promising benchmark results.
Paper: arxiv.org/abs/2609.02734 Details: teapawt.net/dimativatorr/status/20…
1) Muon works because it updates a matrix as a matrix. LoRA splits that matrix into A and B, then most optimizers update the factors independently. The induced step on BA can point somewhere quite different. We built LoRA-TSD to put Muon's geometry back where it belongs.
4
459
✈️ Last week, BRAIn Lab attended UAI 2026 in Amsterdam, presenting our paper "Markovian Compression: Looking to the Past Helps Accelerate the Future" arxiv.org/abs/2601.05398 – at Amsterdam, The Netherlands
3
5
176
🌏 ICML 2026 in Seoul brought together thousands of researchers, engineers, and enthusiasts from around the world. Sharing the impressions of three of our senior researchers @shkodnik1917, Gleb Molodtsov and Nail Bashirov who attended the conference.
























