Max Kleiman-Weiner
932 posts
Max Kleiman-Weiner
@maxhkw
professor @UW scientist @GoogleDeepMind. computational cognitive models of social minds and machines. priors: PhD @MIT founder @diffeo @CSM_ai
- Data driven "super resolution" with deep learning. Might need to curate Ryan Gosling out of the dataset... (from petapixel.com/2020/08/17/gig…) Original ENHANCE
- Before mice open their eyes, their retinas are filled with "synthetic data", spontaneous neural activity that simulates the optic flow pattern generated by forward self-motion. This "pretraining" has causal benefits for motion detection behavior. science.org/doi/10.1126/sc…
00:00 - I’m recruiting PhD students to join the Computational Minds and Machines Lab at the University of Washington in Seattle! Join us to work at the intersection of computational cognitive science and AI with a broad focus on social intelligence. (Please reshare!)
- The paper for PaLM (Google's new language model trained with 540B parameters on 780B tokens) breaks down the team structure for a mega scale machine learning project. storage.googleapis.com/pathways-langu…
- Humans switch from model-free to model-based reinforcement learning only after receiving instructions about how an abstract task works. Social learning is a hidden default in many "learning from experience" experiments
- New work from Kevin Ellis is a breakthrough for learning to synthesize programs. DreamCoder learns and extends a DSL and uses it to solve new problems faster. Rediscovers human-like languages: physical laws, vector algebra & functional programming: arxiv.org/pdf/2006.08381…
- Excited to show off the world model we've built! World models let us predict what we cannot observe directly, plan by simulating possibilities, and imagine new concepts and ideas. We started Common Sense Machines to build AI that understands the world as well as we do.We have built CommonSim-1, a neural simulation engine controlled by images, actions and text. This is a new era in generative AI where simulators will grow and adapt with experience. Read on for more and sign up for early access to apps/API/open-source: csm.ai/commonsim-1-ge…
00:00 - Anyone know how the AVs and Teslas are doing in the Bay Area under today's distribution shift in sensor inputs? Do the perception systems need to be retrained with a sepia filter?
- Cognitive development milestones over the first 18 months of human life. A Gantt chart for building human-like AI from the DARPA Machine Common Sense challenge to build core cognition of objects, agents, and places into machines: darpa.mil/news-events/20…
- Our new work “Too many cooks: Coordinating multi-agent collaboration through inverse planning” won the computational modeling prize for higher cognition at #CogSci2020! Project led by Sarah Wu and @rose_e_wang who will talk on 8/1 at 2:40pm ET. Paper: arxiv.org/abs/2003.11778
GIF - Cool to compare DallE2 with Latent Diffusion LAION-400M on the same prompt. I tried "A plush toy robot sitting against a yellow wall" (left) Dalle (right) Latent Diffusion
- Josh Tenenbaum on scaling up vs growing up and the path to human-like reasoning #NeurIPS2024
- A social path to human-like artificial intelligence Interesting theory paper on how social interaction across multiple scales creates novel data. A nice follow-up to the auto-curriculum ideas from the same authors.
