14 accepted papers
Future advanced AI systems may learn sophisticated strategies through reinforcement learning (RL) that humans cannot understand well enough to safely evaluate. We propose a training method which avoids agents learning undesired multi-step plans that receive high reward (multi-step "reward hacks") ev…
Causal models of agents have been used to analyse the safety aspects of machine learning systems. But identifying agents is non-trivial – often the causal model is just assumed by the modeller without much justification – and modelling failures can lead to mistakes in the safety analysis. This paper…
We investigate the benefit of treating all the parameters in a Bayesian neural network stochastically and find compelling theoretical and empirical evidence that this standard construction may be unnecessary. To this end, we prove that expressive predictive distributions require only small amounts o…
Information-theoretic approaches to active learning have traditionally focused on maximising the information gathered about the model parameters, most commonly by optimising the BALD score. We highlight that this can be suboptimal from the perspective of predictive performance. For example, BALD lac…
We introduce a method to measure uncertainty in large language models. For tasks like question answering, it is essential to know when we can trust the natural language outputs of foundation models. We show that measuring uncertainty in natural language is challenging because of "semantic equivalenc…
We show how to "compile" human-readable programs into standard decoder-only transformer models. Our compiler, Tracr, generates models with known structure. This structure can be used to design experiments. For example, we use it to study "superposition" in transformers that execute multi-step algori…
We propose Active Surrogate Estimators (ASEs), a new method for label-efficient model evaluation. Evaluating model performance is a challenging and important problem when labels are expensive. ASEs address this active testing problem using a surrogate-based estimation approach that interpolates the…
We present a general framework for training safe agents whose naive incentives are unsafe. As an example, manipulative or deceptive behaviour can improve rewards but should be avoided. Most approaches fail here: agents maximize expected return by any means necessary. We formally describe settings wi…
Training on web-scale data can take months. But much computation and time is wasted on redundant and noisy points that are already learnt or not learnable. To accelerate training, we introduce Reducible Holdout Loss Selection (RHO-LOSS), a simple but principled technique which selects approximately…
Pruning neural networks at initialization would enable us to find sparse models that retain the accuracy of the original network while consuming fewer computational resources for training and inference. However, current methods are insufficient to enable this optimization and lead to a large degrada…
We introduce a new framework for sample-efficient model evaluation that we call active testing. While approaches like active learning reduce the number of labels needed for model training, existing literature largely ignores the cost of labeling test data, typically unrealistically assuming large te…
Active learning is a powerful tool when labelling data is expensive, but it introduces a bias because the training data no longer follows the population distribution. We formalize this bias and investigate the situations in which it can be harmful and sometimes even helpful. We further introduce nov…
We challenge the longstanding assumption that the mean-field approximation for variational inference in Bayesian neural networks is severely restrictive, and show this is not the case in deep networks. We prove several results indicating that deep mean-field variational weight posteriors can induce…
We propose Radial Bayesian Neural Networks (BNNs): a variational approximate posterior for BNNs which scales well to large models. Unlike scalable Bayesian deep learning methods like deep ensembles that have discrete support (assign exactly zero probability almost everywhere in weight-space) Radial…