Guy Van den Broeck
@guyvdb
Followers
4K
Following
5K
Media
0
Statuses
41
Professor of Computer Science and Samueli Fellow at UCLA @UCLAComSci; Scientist at @RelationalAI; working on Artificial Intelligence
Los Angeles, CA
Joined April 2008
What happens when we compress the KV cache of prompts with multiple instructions? 🤔 Existing compression methods can lead to some instructions being ignored. 🙀 We propose simple changes to KV cache eviction that fix this problem alongside other pitfalls to be aware of. 💯
2
2
15
Plan autoregressively, denoise in parallel!
"An hour of planning can save you 10 hours of doing." ✨📝 Planned Diffusion 📝 ✨ makes a plan before parallel dLLM generation. Planned Diffusion runs 1.2-1.8× faster than autoregressive and an order of magnitude faster than diffusion, while staying within 0.9–5% AR quality.
0
2
5
Diffusion 🤝 Autoregressive Fast high-quality generation
"An hour of planning can save you 10 hours of doing." ✨📝 Planned Diffusion 📝 ✨ makes a plan before parallel dLLM generation. Planned Diffusion runs 1.2-1.8× faster than autoregressive and an order of magnitude faster than diffusion, while staying within 0.9–5% AR quality.
0
2
2
"An hour of planning can save you 10 hours of doing." ✨📝 Planned Diffusion 📝 ✨ makes a plan before parallel dLLM generation. Planned Diffusion runs 1.2-1.8× faster than autoregressive and an order of magnitude faster than diffusion, while staying within 0.9–5% AR quality.
7
45
308
🔦Adaptive Parallel Decoding (APD) has been accepted as a spotlight paper at @NeurIPSConf ! I thank my collaborators, reviewers, and program organizers for this honor. A thread for those interested 🧵 (1/n)
11
22
168
@e_giunchiglia @guyvdb How can reverend Bayes help us to incorporate constraints? With NeSy of course 👀 With applications in non-toxic LLM generation and safe AI driving! @guyvdb
1
1
6
@e_giunchiglia Now, @guyvdb is giving the opening keynote arguing why symbolic AI is still relevant in the age of LLMs... With the help of Shrek!
1
5
12
@e_giunchiglia @guyvdb Behind all of these very nice methods are one central trick... Circuits! ➕✖️ These are tractable generative neural networks 😍
1
2
6
1
1
2
#Nesy2025 schedule is out! Keynote speakers include @guyvdb @tkipf @dlmcguinness and @GaryMarcus. I'm looking forward to seeing what everyone's cooking and meeting everyone! @nesyconf
https://t.co/5AS6shaNoH
2025.nesyconf.org
19th International Conference on Neurosymbolic Learning and Reasoning (NeSy 2025, 8-10 September 2025, Santa Cruz, CA, USA)
1
2
7
Wish LM could plan—not just guess the next word? TRACE lets LM see all endings before each move. – Global control at inference time – Tractable lookahead via an HMM LM-proxy – Linear classifier per constraint Outperform RL, DPO, FUDGE—at just +20% decoding over base LM. #ICML2025
10
1
7
🚨 First Call for Participation – NeSy 2025 📍 Sept 8–10 | Santa Cruz, CA Join the longest-running conference on neurosymbolic AI! Our keynote speakers: @guyvdb , @tkipf , @dlmcguinness , @GaryMarcus More info 👇
1
10
19
Announcing the 2025 IJCAI Computers and Thought Award winner ✨Aditya Grover @adityagrover_, @InceptionAILabs @UCLA. Dr. Grover is honored for uniting deep generative models, representation learning & RL to advance scientific reasoning. Congratulations! https://t.co/Z3xESFizpi
3
13
72
Accelerating Diffusion LLMs via Adaptive Parallel Decoding "We therefore introduce adaptive parallel decoding (APD), a novel method that dynamically adjusts the number of tokens sampled in parallel." "Notably, Dream with ADP surpasses the speed of autoregressive Qwen 7B and
7
17
167