Deep Learning
Google DeepMind Hires Former CTO of Boston Dynamics as the Company Pushes Deeper Into Robotics
DeepMind's chief says he envisions Gemini as an operating system for physical robots. The company has hired Aaron Saunders to help make that a reality. Google DeepMind has hired the former Chief Technology Officer of Boston Dynamics as the company pushes deeper into robotics. Aaron Saunders, who is partly responsible for giving the world backflipping and dancing machines, joined as the VP of hardware engineering earlier this month. The hire is a key part of CEO Demis Hassabis' vision for Gemini to become a sort of robot operating system, similar to how Google supplies its Android software to an array of smartphone manufacturers.
Exploiting the Replay Memory Before Exploring the Environment: Enhancing Reinforcement Learning Through Empirical MDP Iteration
Reinforcement learning (RL) algorithms are typically based on optimizing a Markov Decision Process (MDP) using the optimal Bellman equation.
The Zelos-450 Pellet Grill Has Features Missing on Grills Triple Its Price
An AI-Enabled Pellet Grill Is a Dumb Idea. Brisk It's Zelos-450 boasts gimmicky generative AI. Grills don't need AI, but you might need an AI grill. When it debuted at CES early this year, the Brisk It Zelos-450 was advertised as being the first grill with generative AI. That AI comes in the form of Vera, an in-app feature that "creates customized recipes based on your input and then sets the grill's temperature to cook them.
self-supervised extraction and projection of generalized velocity signals
Example of an abstract, non-spatial domain called'Stretchy Blob', a 2D Gaussian that can either stretch or shrink along two axes.
990641d09f71bcee0060a8f1704ab8e2-Paper-Conference.pdf
Due to the recurrent structure of RNN, the long information propagation path poses limitations in capturing long-term dependencies, gradient explosion/vanishing issues, and inefficient sequential execution. Based on this, we propose a novel paradigm called Parallel Gated Network (PGN) as the new successor to RNN.
Neural Foundations of Mental Simulation: Future Prediction of Latent Representations on Dynamic Scenes
We find that "scale is not all you need", and that
Direct Preference-based Policy Optimization without Reward Modeling
Instead, we propose a PbRL algorithm that directly learns from preference without requiring any reward modeling.
Leveraging Separated World Model for Exploration in Visually Distracted Environments
Despite its prevalence in real-world environments, this challenge has received limited attention in unsupervised reinforcement learning.
One-Layer Transformer Provably Learns One-Nearest Neighbor In Context
Transformers have achieved great success in recent years. Interestingly, transformers have shown particularly strong in-context learning capability - even without fine-tuning, they are still able to solve unseen tasks well purely based on task-specific prompts.
Normalization Layers Are All That Sharpness-Aware Minimization Needs Maximilian Müller University of Tübingen and Tübingen AI Center
In this work we show that perturbing only the affine normalization parameters (typically comprising 0.1% of the total parameters) in the adversarial step of SAM can outperform perturbing all of the parameters.