Tim Pearce

Tim Pearce, Tabish Rashid, Anssi Kanervisto et al.

Diffusion models have emerged as powerful generative models in the text-to-image domain. This paper studies their application as observation-to-action models for imitating human behaviour in sequential environments. Human behaviour is stochastic and multimodal, with structured correlations between action dimensions. Meanwhile, standard modelling choices in behaviour cloning are limited in their expressiveness and may introduce bias into the cloned policy. We begin by pointing out the limitations of these choices. We then propose that diffusion models are an excellent fit for imitating human behaviour, since they learn an expressive distribution over the joint action space. We introduce several innovations to make diffusion models suitable for sequential environments; designing suitable architectures, investigating the role of guidance, and developing reliable sampling strategies. Experimentally, diffusion models closely match human demonstrations in a simulated robotic control task and a modern 3D gaming environment.

Fanqi Lin, Shiyu Huang, Tim Pearce et al.

Multi-agent football poses an unsolved challenge in AI research. Existing work has focused on tackling simplified scenarios of the game, or else leveraging expert demonstrations. In this paper, we develop a multi-agent system to play the full 11 vs. 11 game mode, without demonstrations. This game mode contains aspects that present major challenges to modern reinforcement learning algorithms; multi-agent coordination, long-term planning, and non-transitivity. To address these challenges, we present TiZero; a self-evolving, multi-agent system that learns from scratch. TiZero introduces several innovations, including adaptive curriculum learning, a novel self-play strategy, and an objective that optimizes the policies of multiple agents jointly. Experimentally, it outperforms previous systems by a large margin on the Google Research Football environment, increasing win rates by over 30%. To demonstrate the generality of TiZero's innovations, they are assessed on several environments beyond football; Overcooked, Multi-agent Particle-Environment, Tic-Tac-Toe and Connect-Four.

Tim Pearce, Jong-Hyeon Jeong, Yichen Jia et al.

This paper considers doing quantile regression on censored data using neural networks (NNs). This adds to the survival analysis toolkit by allowing direct prediction of the target variable, along with a distribution-free characterisation of uncertainty, using a flexible function approximator. We begin by showing how an algorithm popular in linear models can be applied to NNs. However, the resulting procedure is inefficient, requiring sequential optimisation of an individual NN at each desired quantile. Our major contribution is a novel algorithm that simultaneously optimises a grid of quantiles output by a single NN. To offer theoretical insight into our algorithm, we show firstly that it can be interpreted as a form of expectation-maximisation, and secondly that it exhibits a desirable `self-correcting' property. Experimentally, the algorithm produces quantiles that are better calibrated than existing methods on 10 out of 12 real datasets.

Wentse Chen, Shiyu Huang, Yuan Chiang et al.

Most reinforcement learning algorithms seek a single optimal strategy that solves a given task. However, it can often be valuable to learn a diverse set of solutions, for instance, to make an agent's interaction with users more engaging, or improve the robustness of a policy to an unexpected perturbance. We propose Diversity-Guided Policy Optimization (DGPO), an on-policy algorithm that discovers multiple strategies for solving a given task. Unlike prior work, it achieves this with a shared policy network trained over a single run. Specifically, we design an intrinsic reward based on an information-theoretic diversity objective. Our final objective alternately constraints on the diversity of the strategies and on the extrinsic reward. We solve the constrained optimization problem by casting it as a probabilistic inference task and use policy iteration to maximize the derived lower bound. Experimental results show that our method efficiently discovers diverse strategies in a wide variety of reinforcement learning tasks. Compared to baseline methods, DGPO achieves comparable rewards, while discovering more diverse strategies, and often with better sample efficiency.

Stephen Mak, Liming Xu, Tim Pearce et al.

Collaborative vehicle routing occurs when carriers collaborate through sharing their transportation requests and performing transportation requests on behalf of each other. This achieves economies of scale, thus reducing cost, greenhouse gas emissions and road congestion. But which carrier should partner with whom, and how much should each carrier be compensated? Traditional game theoretic solution concepts are expensive to calculate as the characteristic function scales exponentially with the number of agents. This would require solving the vehicle routing problem (NP-hard) an exponential number of times. We therefore propose to model this problem as a coalitional bargaining game solved using deep multi-agent reinforcement learning, where - crucially - agents are not given access to the characteristic function. Instead, we implicitly reason about the characteristic function; thus, when deployed in production, we only need to evaluate the expensive post-collaboration vehicle routing problem once. Our contribution is that we are the first to consider both the route allocation problem and gain sharing problem simultaneously - without access to the expensive characteristic function. Through decentralised machine learning, our agents bargain with each other and agree to outcomes that correlate well with the Shapley value - a fair profit allocation mechanism. Importantly, we are able to achieve a reduction in run-time of 88%.

Stephen Mak, Liming Xu, Tim Pearce et al.

Collaborative Vehicle Routing is where delivery companies cooperate by sharing their delivery information and performing delivery requests on behalf of each other. This achieves economies of scale and thus reduces cost, greenhouse gas emissions, and road congestion. But which company should partner with whom, and how much should each company be compensated? Traditional game theoretic solution concepts, such as the Shapley value or nucleolus, are difficult to calculate for the real-world problem of Collaborative Vehicle Routing due to the characteristic function scaling exponentially with the number of agents. This would require solving the Vehicle Routing Problem (an NP-Hard problem) an exponential number of times. We therefore propose to model this problem as a coalitional bargaining game where - crucially - agents are not given access to the characteristic function. Instead, we implicitly reason about the characteristic function, and thus eliminate the need to evaluate the VRP an exponential number of times - we only need to evaluate it once. Our contribution is that our decentralised approach is both scalable and considers the self-interested nature of companies. The agents learn using a modified Independent Proximal Policy Optimisation. Our RL agents outperform a strong heuristic bot. The agents correctly identify the optimal coalitions 79% of the time with an average optimality gap of 4.2% and reduction in run-time of 62%.

Junliang Guo, Yang Ye, Tianyu He et al.

World modeling is a crucial task for enabling intelligent agents to effectively interact with humans and operate in dynamic environments. In this work, we propose MineWorld, a real-time interactive world model on Minecraft, an open-ended sandbox game which has been utilized as a common testbed for world modeling. MineWorld is driven by a visual-action autoregressive Transformer, which takes paired game scenes and corresponding actions as input, and generates consequent new scenes following the actions. Specifically, by transforming visual game scenes and actions into discrete token ids with an image tokenizer and an action tokenizer correspondingly, we consist the model input with the concatenation of the two kinds of ids interleaved. The model is then trained with next token prediction to learn rich representations of game states as well as the conditions between states and actions simultaneously. In inference, we develop a novel parallel decoding algorithm that predicts the spatial redundant tokens in each frame at the same time, letting models in different scales generate $4$ to $7$ frames per second and enabling real-time interactions with game players. In evaluation, we propose new metrics to assess not only visual quality but also the action following capacity when generating new scenes, which is crucial for a world model. Our comprehensive evaluation shows the efficacy of MineWorld, outperforming SoTA open-sourced diffusion based world models significantly. The code and model have been released.

Samuel Garcin, Thomas Walker, Steven McDonagh et al.

Interactive world models continually generate video by responding to a user's actions, enabling open-ended generation capabilities. However, existing models typically lack a 3D representation of the environment, meaning 3D consistency must be implicitly learned from data, and spatial memory is restricted to limited temporal context windows. This results in an unrealistic user experience and presents significant obstacles to down-stream tasks such as training agents. To address this, we present PERSIST, a new paradigm of world model which simulates the evolution of a latent 3D scene: environment, camera, and renderer. This allows us to synthesize new frames with persistent spatial memory and consistent geometry. Both quantitative metrics and a qualitative user study show substantial improvements in spatial memory, 3D consistency, and long-horizon stability over existing methods, enabling coherent, evolving 3D worlds. We further demonstrate novel capabilities, including synthesising diverse 3D environments from a single image, as well as enabling fine-grained, geometry-aware control over generated experiences by supporting environment editing and specification directly in 3D space. Project page: https://francelico.github.io/persist.github.io

Hangxing Wei, Xiaoyu Chen, Chuheng Zhang et al.

Latent action learning infers pseudo-action labels from visual transitions, providing an approach to leverage internet-scale video for embodied AI. However, most methods learn latent actions without structural priors that encode the additive, compositional structure of physical motion. As a result, latents often entangle irrelevant scene details or information about future observations with true state changes and miscalibrate motion magnitude. We introduce Additively Compositional Latent Action Model (AC-LAM), which enforces scene-wise additive composition structure over short horizons on the latent action space. These AC constraints encourage simple algebraic structure in the latent action space~(identity, inverse, cycle consistency) and suppress information that does not compose additively. Empirically, AC-LAM learns more structured, motion-specific, and displacement-calibrated latent actions and provides stronger supervision for downstream policy learning, outperforming state-of-the-art LAMs across simulated and real-world tabletop tasks.

Eloi Alonso, Adam Jelley, Vincent Micheli et al.

World models constitute a promising approach for training reinforcement learning agents in a safe and sample-efficient manner. Recent world models predominantly operate on sequences of discrete latent variables to model environment dynamics. However, this compression into a compact discrete representation may ignore visual details that are important for reinforcement learning. Concurrently, diffusion models have become a dominant approach for image generation, challenging well-established methods modeling discrete latents. Motivated by this paradigm shift, we introduce DIAMOND (DIffusion As a Model Of eNvironment Dreams), a reinforcement learning agent trained in a diffusion world model. We analyze the key design choices that are required to make diffusion suitable for world modeling, and demonstrate how improved visual details can lead to improved agent performance. DIAMOND achieves a mean human normalized score of 1.46 on the competitive Atari 100k benchmark; a new best for agents trained entirely within a world model. We further demonstrate that DIAMOND's diffusion world model can stand alone as an interactive neural game engine by training on static Counter-Strike: Global Offensive gameplay. To foster future research on diffusion for world modeling, we release our code, agents, videos and playable world models at https://diamond-wm.github.io.

Russell Tsuchida, Tim Pearce, Chris van der Heide et al.

Analysing and computing with Gaussian processes arising from infinitely wide neural networks has recently seen a resurgence in popularity. Despite this, many explicit covariance functions of networks with activation functions used in modern networks remain unknown. Furthermore, while the kernels of deep networks can be computed iteratively, theoretical understanding of deep kernels is lacking, particularly with respect to fixed-point dynamics. Firstly, we derive the covariance functions of multi-layer perceptrons (MLPs) with exponential linear units (ELU) and Gaussian error linear units (GELU) and evaluate the performance of the limiting Gaussian processes on some benchmarks. Secondly, and more generally, we analyse the fixed-point dynamics of iterated kernels corresponding to a broad range of activation functions. We find that unlike some previously studied neural network kernels, these new kernels exhibit non-trivial fixed-point dynamics which are mirrored in finite-width neural networks. The fixed point behaviour present in some networks explains a mechanism for implicit regularisation in overparameterised deep models. Our results relate to both the static iid parameter conjugate kernel and the dynamic neural tangent kernel constructions. Software at github.com/RussellTsuchida/ELU_GELU_kernels.

Tim Pearce, Tabish Rashid, Dave Bignell et al.

The performance of embodied agents has been shown to improve by increasing model parameters, dataset size, and compute. This has been demonstrated in domains from robotics to video games, when generative learning objectives on offline datasets (pre-training) are used to model an agent's behavior (imitation learning) or their environment (world modeling). This paper characterizes the role of scale in these tasks more precisely. Going beyond the simple intuition that `bigger is better', we show that the same types of power laws found in language modeling also arise in world modeling and imitation learning (e.g. between loss and optimal model size). However, the coefficients of these laws are heavily influenced by the tokenizer, task \& architecture -- this has important implications on the optimal sizing of models and data.

Tianjiao Luo, Tim Pearce, Huayu Chen et al.

Generative Adversarial Imitation Learning (GAIL) trains a generative policy to mimic a demonstrator. It uses on-policy Reinforcement Learning (RL) to optimize a reward signal derived from a GAN-like discriminator. A major drawback of GAIL is its training instability - it inherits the complex training dynamics of GANs, and the distribution shift introduced by RL. This can cause oscillations during training, harming its sample efficiency and final policy performance. Recent work has shown that control theory can help with the convergence of a GAN's training. This paper extends this line of work, conducting a control-theoretic analysis of GAIL and deriving a novel controller that not only pushes GAIL to the desired equilibrium but also achieves asymptotic stability in a 'one-step' setting. Based on this, we propose a practical algorithm 'Controlled-GAIL' (C-GAIL). On MuJoCo tasks, our controlled variant is able to speed up the rate of convergence, reduce the range of oscillation and match the expert's distribution more closely both for vanilla GAIL and GAIL-DAC.

Chuheng Zhang, Tim Pearce, Pushi Zhang et al. · tsinghua

Latent action models (LAMs) aim to learn action-relevant changes from unlabeled videos by compressing changes between frames as latents. However, differences between video frames can be caused by controllable changes as well as exogenous noise, leading to an important concern -- do latents capture the changes caused by actions or irrelevant noise? This paper studies this issue analytically, presenting a linear model that encapsulates the essence of LAM learning, while being tractable.This provides several insights, including connections between LAM and principal component analysis (PCA), desiderata of the data-generating policy, and justification of strategies to encourage learning controllable changes using data augmentation, data cleaning, and auxiliary action-prediction. We also provide illustrative results based on numerical simulation, shedding light on the specific structure of observations, actions, and noise in data that influence LAM learning.

Yang Ye, Junliang Guo, Haoyu Wu et al.

Autoregressive Transformer models have demonstrated impressive performance in video generation, but their sequential token-by-token decoding process poses a major bottleneck, particularly for long videos represented by tens of thousands of tokens. In this paper, we propose Diagonal Decoding (DiagD), a training-free inference acceleration algorithm for autoregressively pre-trained models that exploits spatial and temporal correlations in videos. Our method generates tokens along diagonal paths in the spatial-temporal token grid, enabling parallel decoding within each frame as well as partially overlapping across consecutive frames. The proposed algorithm is versatile and adaptive to various generative models and tasks, while providing flexible control over the trade-off between inference speed and visual quality. Furthermore, we propose a cost-effective finetuning strategy that aligns the attention patterns of the model with our decoding order, further mitigating the training-inference gap on small-scale models. Experiments on multiple autoregressive video generation models and datasets demonstrate that DiagD achieves up to $10\times$ speedup compared to naive sequential decoding, while maintaining comparable visual fidelity.

Tim Pearce, Jinyeop Song

Kaplan et al. [2020] (`Kaplan') and Hoffmann et al. [2022] (`Chinchilla') studied the scaling behavior of transformers trained on next-token language prediction. These studies produced different estimates for how the number of parameters ($N$) and training tokens ($D$) should be set to achieve the lowest possible loss for a given compute budget ($C$). Kaplan: $N_\text{optimal} \propto C^{0.73}$, Chinchilla: $N_\text{optimal} \propto C^{0.50}$. This paper finds that much of this discrepancy can be attributed to Kaplan counting non-embedding rather than total parameters, combined with their analysis being performed at small scale. Simulating the Chinchilla study under these conditions produces biased scaling coefficients close to Kaplan's. Hence, this paper reaffirms Chinchilla's scaling coefficients, by explaining the primary cause of Kaplan's original overestimation. As a second contribution, the paper explains differences in the reported relationships between loss and compute. These findings lead us to recommend that future scaling studies use total parameters and compute.

Bang Xiang Yong, Tim Pearce, Alexandra Brintrup

After an autoencoder (AE) has learnt to reconstruct one dataset, it might be expected that the likelihood on an out-of-distribution (OOD) input would be low. This has been studied as an approach to detect OOD inputs. Recent work showed this intuitive approach can fail for the dataset pairs FashionMNIST vs MNIST. This paper suggests this is due to the use of Bernoulli likelihood and analyses why this is the case, proposing two fixes: 1) Compute the uncertainty of likelihood estimate by using a Bayesian version of the AE. 2) Use alternative distributions to model the likelihood.

Tim Pearce, Alexandra Brintrup, Jun Zhu

It is often remarked that neural networks fail to increase their uncertainty when predicting on data far from the training distribution. Yet naively using softmax confidence as a proxy for uncertainty achieves modest success in tasks exclusively testing for this, e.g., out-of-distribution (OOD) detection. This paper investigates this contradiction, identifying two implicit biases that do encourage softmax confidence to correlate with epistemic uncertainty: 1) Approximately optimal decision boundary structure, and 2) Filtering effects of deep networks. It describes why low-dimensional intuitions about softmax confidence are misleading. Diagnostic experiments quantify reasons softmax confidence can fail, finding that extrapolations are less to blame than overlap between training and OOD data in final-layer representations. Pre-trained/fine-tuned networks reduce this overlap.

Tim Pearce, Jun Zhu

This paper describes an AI agent that plays the popular first-person-shooter (FPS) video game `Counter-Strike; Global Offensive' (CSGO) from pixel input. The agent, a deep neural network, matches the performance of the medium difficulty built-in AI on the deathmatch game mode, whilst adopting a humanlike play style. Unlike much prior work in games, no API is available for CSGO, so algorithms must train and run in real-time. This limits the quantity of on-policy data that can be generated, precluding many reinforcement learning algorithms. Our solution uses behavioural cloning - training on a large noisy dataset scraped from human play on online servers (4 million frames, comparable in size to ImageNet), and a smaller dataset of high-quality expert demonstrations. This scale is an order of magnitude larger than prior work on imitation learning in FPS games.

Tim Pearce, Andrew Y. K. Foong, Alexandra Brintrup

Selection of an architectural prior well suited to a task (e.g. convolutions for image data) is crucial to the success of deep neural networks (NNs). Conversely, the weight priors within these architectures are typically left vague, e.g.~independent Gaussian distributions, which has led to debate over the utility of Bayesian deep learning. This paper explores the benefits of adding structure to weight priors. It initially considers first-layer filters of a convolutional NN, designing a prior based on random Gabor filters. Second, it considers adding structure to the prior of final-layer weights by estimating how each hidden feature relates to each class. Empirical results suggest that these structured weight priors lead to more meaningful functional priors for image data. This contributes to the ongoing discussion on the importance of weight priors.

Tim Pearce, Russell Tsuchida, Mohamed Zaki et al.

A simple, flexible approach to creating expressive priors in Gaussian process (GP) models makes new kernels from a combination of basic kernels, e.g. summing a periodic and linear kernel can capture seasonal variation with a long term trend. Despite a well-studied link between GPs and Bayesian neural networks (BNNs), the BNN analogue of this has not yet been explored. This paper derives BNN architectures mirroring such kernel combinations. Furthermore, it shows how BNNs can produce periodic kernels, which are often useful in this context. These ideas provide a principled approach to designing BNNs that incorporate prior knowledge about a function. We showcase the practical value of these ideas with illustrative experiments in supervised and reinforcement learning settings.

Tim Pearce, Mohamed Zaki, Andy Neely

Ensembles of neural networks (NNs) have long been used to estimate predictive uncertainty; a small number of NNs are trained from different initialisations and sometimes on differing versions of the dataset. The variance of the ensemble's predictions is interpreted as its epistemic uncertainty. The appeal of ensembling stems from being a collection of regular NNs - this makes them both scalable and easily implementable. They have achieved strong empirical results in recent years, often presented as a practical alternative to more costly Bayesian NNs (BNNs). The departure from Bayesian methodology is of concern since the Bayesian framework provides a principled, widely-accepted approach to handling uncertainty. In this extended abstract we derive and implement a modified NN ensembling scheme, which provides a consistent estimator of the Bayesian posterior in wide NNs - regularising parameters about values drawn from a prior distribution.

Tim Pearce, Felix Leibfried, Alexandra Brintrup et al.

Understanding the uncertainty of a neural network's (NN) predictions is essential for many purposes. The Bayesian framework provides a principled approach to this, however applying it to NNs is challenging due to large numbers of parameters and data. Ensembling NNs provides an easily implementable, scalable method for uncertainty quantification, however, it has been criticised for not being Bayesian. This work proposes one modification to the usual process that we argue does result in approximate Bayesian inference; regularising parameters about values drawn from a distribution which can be set equal to the prior. A theoretical analysis of the procedure in a simplified setting suggests the recovered posterior is centred correctly but tends to have an underestimated marginal variance, and overestimated correlation. However, two conditions can lead to exact recovery. We argue that these conditions are partially present in NNs. Empirical evaluations demonstrate it has an advantage over standard ensembling, and is competitive with variational methods.

Tim Pearce, Nicolas Anastassacos, Mohamed Zaki et al.

The use of ensembles of neural networks (NNs) for the quantification of predictive uncertainty is widespread. However, the current justification is intuitive rather than analytical. This work proposes one minor modification to the normal ensembling methodology, which we prove allows the ensemble to perform Bayesian inference, hence converging to the corresponding Gaussian Process as both the total number of NNs, and the size of each, tend to infinity. This working paper provides early-stage results in a reinforcement learning setting, analysing the practicality of the technique for an ensemble of small, finite number. Using the uncertainty estimates produced by anchored ensembles to govern the exploration-exploitation process results in steadier, more stable learning.

Tim Pearce, Mohamed Zaki, Alexandra Brintrup et al.

This paper considers the generation of prediction intervals (PIs) by neural networks for quantifying uncertainty in regression tasks. It is axiomatic that high-quality PIs should be as narrow as possible, whilst capturing a specified portion of data. We derive a loss function directly from this axiom that requires no distributional assumption. We show how its form derives from a likelihood principle, that it can be used with gradient descent, and that model uncertainty is accounted for in ensembled form. Benchmark experiments show the method outperforms current state-of-the-art uncertainty quantification methods, reducing average PI width by over 10%.