🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Azerbaijan, Turkey, and countries supporting them, by definition, are committing acts of terrorism. This itself is not surprising, as both countries are led by dictators having medieval thinking. The real surprise is the tolerance of the international community. #ՀԱՂԹԵԼՈԻԵՆՔ

Hamiltonian Dynamics with Non-Newtonian Momentum for Rapid Sampling deepai.org/publication/ha… by Greg Ver Steeg and Aram Galstyan #EnergybasedModels #NeuralNetwork

Last week Artificial Intelligence @ KAUST hosted the 2nd iteration of the rising stars symposium 2023 that is geared towards young researchers, who have recently published significant work at leading AI venues. It was a great opportunity for attendees to discuss and exchange exciting AI research ideas

We are organizing a Math conference in Armenia from July 3 to July 8. If you wish to discover a beautiful country and listen to great talks, please check the website mathconf.sci.am.

If you want to spend nice holidays, learn some exciting topics from Stat & ML, and discover a country, here is a perfect opportunity: summer school in Stat & ML mathschool.ysu.am/slt2023

We're organizing the first summer course on LLMs in Armenia this year! We'll cover the foundations of LLMs from first principles through lectures from a great lineup of speakers and hands-on practice sessions. If interested, reach out directly or go to armllm.github.io/2024/.

Excited to share our new paper (NeurIPS '24) on stacking and its inductive biases! TLDR: Stacking not only improves training efficiency (if done right), but significantly improves downstream tasks that require *reasoning*, at similar perplexity. 1/n arxiv.org/pdf/2409.19044

State space models have struggled to learn to do things like copying and associative recall 🟢 -- things that self-attention learns easily 🟠... But it turns out we just needed to change SSM initialization a bit 🔵. Our init helps a lot, and even makes state space layers *look*

[LG] A Little Help Goes a Long Way: Efficient LLM Training by Leveraging Small LMs A S Rawat, V Sadhanala, A Rostamizadeh, A Chakrabarti... [Google Research] (2024) arxiv.org/abs/2410.18779

🚀 Excited to share our latest research Google DeepMind on ♻️Recursive Transformers! We make smaller LMs by "sharing parameters" across layers. A novel serving paradigm, ✨Continuous Depth-wise Batching, with 🏃Early-Exiting could significantly boost their decoding speed! 🧵👇

Our recent work explores "Chain-of-Thought with Continuous Tokens (CoT2),” to facilitate language model reasoning with continuous mixtures of discrete tokens. We introduce optimization and RL methods for CoT2, paving the way for more expressive inference.🧵