Training Transformers with Enforced Lipschitz Constants

Neural networks are often highly sensitive to input and weight perturbations. This sensitivity has been linked to pathologies such as vulnerability to adversarial examples, divergent training, and overfitting. To combat these problems, past research has looked at building neural networks entirely from Lipschitz components. However, these techniques have not matured to the point where researchers have trained a modern architecture such as a transformer with a Lipschitz certificate enforced beyond initialization. To explore this gap, we begin by developing and benchmarking novel, computationally-efficient tools for maintaining norm-constrained weight matrices. Applying these tools, we are able to train transformer models with Lipschitz bounds enforced throughout training. We find that optimizer dynamics matter: switching from AdamW to Muon improves standard methods -- weight decay and spectral normalization -- allowing models to reach equal performance with a lower Lipschitz bound. Inspired by Muon's update having a fixed spectral norm, we co-design a weight constraint method that improves the Lipschitz vs. performance tradeoff on MLPs and 2M parameter transformers. Our 2-Lipschitz transformer on Shakespeare text reaches validation accuracy 60%. Scaling to 145M parameters, our 10-Lipschitz transformer reaches 21% accuracy on internet text. However, to match the NanoGPT baseline validation accuracy of 39.4%, our Lipschitz upper bound increases to 10^264. Nonetheless, our Lipschitz transformers train without stability measures such as layer norm, QK norm, and logit tanh softcapping.

Improving World Models using Deep Supervision with Linear Probes

Developing effective world models is crucial for creating artificial agents that can reason about and navigate complex environments. In this paper, we investigate a deep supervision technique for encouraging the development of a world model in a network trained end-to-end to predict the next observation. While deep supervision has been widely applied for task-specific learning, our focus is on improving the world models. Using an experimental environment based on the Flappy Bird game, where the agent receives only LIDAR measurements as observations, we explore the effect of adding a linear probe component to the network's loss function. This additional term encourages the network to encode a subset of the true underlying world features into its hidden state. Our experiments demonstrate that this supervision technique improves both training and test performance, enhances training stability, and results in more easily decodable world features -- even for those world features which were not included in the training. Furthermore, we observe a reduced distribution drift in networks trained with the linear probe, particularly during high-variability phases of the game (flying between successive pipe encounters). Including the world features loss component roughly corresponded to doubling the model size, suggesting that the linear probe technique is particularly beneficial in compute-limited settings or when aiming to achieve the best performance with smaller models. These findings contribute to our understanding of how to develop more robust and sophisticated world models in artificial agents, paving the way for further advancements in this field.

Project Jenkins: Turning Monkey Neural Data into Robotic Arm Movement, and Back

Project Jenkins explores how neural activity in the brain can be decoded into robotic movement and, conversely, how movement patterns can be used to generate synthetic neural data. Using real neural data recorded from motor and premotor cortex areas of a macaque monkey named Jenkins, we develop models for decoding (converting brain signals into robotic arm movements) and encoding (simulating brain activity corresponding to a given movement). For the interface between the brain simulation and the physical world, we utilized Koch v1.1 leader and follower robotic arms. We developed an interactive web console that allows users to generate synthetic brain data from joystick movements in real time. Our results are a step towards brain-controlled robotics, prosthetics, and enhancing normal motor function. By accurately modeling brain activity, we take a step toward flexible brain-computer interfaces that generalize beyond predefined movements. To support the research community, we provide open source tools for both synthetic data generation and neural decoding, fostering reproducibility and accelerating progress. The project is available at https://www.808robots.com/projects/jenkins