Offline Reinforcement Learning via High-Fidelity Generative Behavior Modeling

In offline reinforcement learning, weighted regression is a common method to ensure the learned policy stays close to the behavior policy and to prevent selecting out-of-sample actions. In this work, we show that due to the limited distributional expressivity of policy models, previous methods might still select unseen actions during training, which deviates from their initial motivation. To address this problem, we adopt a generative approach by decoupling the learned policy into two parts: an expressive generative behavior model and an action evaluation model. The key insight is that such decoupling avoids learning an explicitly parameterized policy model with a closed-form expression. Directly learning the behavior policy allows us to leverage existing advances in generative modeling, such as diffusion-based methods, to model diverse behaviors. As for action evaluation, we combine our method with an in-sample planning technique to further avoid selecting out-of-sample actions and increase computational efficiency. Experimental results on D4RL datasets show that our proposed method achieves competitive or superior performance compared with state-of-the-art offline RL methods, especially in complex tasks such as AntMaze. We also empirically demonstrate that our method can successfully learn from a heterogeneous dataset containing multiple distinctive but similarly successful strategies, whereas previous unimodal policies fail.

All are Worth Words: a ViT Backbone for Score-based Diffusion Models

Vision transformers (ViT) have shown promise in various vision tasks including low-level ones while the U-Net remains dominant in score-based diffusion models. In this paper, we perform a systematical empirical study on the ViT-based architectures in diffusion models. Our results suggest that adding extra long skip connections (like the U-Net) to ViT is crucial to diffusion models. The new ViT architecture, together with other improvements, is referred to as U-ViT. On several popular visual datasets, U-ViT achieves competitive generation results to SOTA U-Net while requiring comparable amount of parameters and computation if not less.

Bi-level Physics-Informed Neural Networks for PDE Constrained Optimization using Broyden's Hypergradients

Deep learning based approaches like Physics-informed neural networks (PINNs) and DeepONets have shown promise on solving PDE constrained optimization (PDECO) problems. However, existing methods are insufficient to handle those PDE constraints that have a complicated or nonlinear dependency on optimization targets. In this paper, we present a novel bi-level optimization framework to resolve the challenge by decoupling the optimization of the targets and constraints. For the inner loop optimization, we adopt PINNs to solve the PDE constraints only. For the outer loop, we design a novel method by using Broyden's method based on the Implicit Function Theorem (IFT), which is efficient and accurate for approximating hypergradients. We further present theoretical explanations and error analysis of the hypergradients computation. Extensive experiments on multiple large-scale and nonlinear PDE constrained optimization problems demonstrate that our method achieves state-of-the-art results compared with strong baselines.

On the Reuse Bias in Off-Policy Reinforcement Learning

Importance sampling (IS) is a popular technique in off-policy evaluation, which re-weights the return of trajectories in the replay buffer to boost sample efficiency. However, training with IS can be unstable and previous attempts to address this issue mainly focus on analyzing the variance of IS. In this paper, we reveal that the instability is also related to a new notion of Reuse Bias of IS -- the bias in off-policy evaluation caused by the reuse of the replay buffer for evaluation and optimization. We theoretically show that the off-policy evaluation and optimization of the current policy with the data from the replay buffer result in an overestimation of the objective, which may cause an erroneous gradient update and degenerate the performance. We further provide a high-probability upper bound of the Reuse Bias, and show that controlling one term of the upper bound can control the Reuse Bias by introducing the concept of stability for off-policy algorithms. Based on these analyses, we finally present a novel Bias-Regularized Importance Sampling (BIRIS) framework along with practical algorithms, which can alleviate the negative impact of the Reuse Bias. Experimental results show that our BIRIS-based methods can significantly improve the sample efficiency on a series of continuous control tasks in MuJoCo.

Regret Analysis for Hierarchical Experts Bandit Problem

We study an extension of standard bandit problem in which there are R layers of experts. Multi-layered experts make selections layer by layer and only the experts in the last layer can play arms. The goal of the learning policy is to minimize the total regret in this hierarchical experts setting. We first analyze the case that total regret grows linearly with the number of layers. Then we focus on the case that all experts are playing Upper Confidence Bound (UCB) strategy and give several sub-linear upper bounds for different circumstances. Finally, we design some experiments to help the regret analysis for the general case of hierarchical UCB structure and show the practical significance of our theoretical results. This article gives many insights about reasonable hierarchical decision structure.

Robust Learning of Deep Time Series Anomaly Detection Models with Contaminated Training Data

Time series anomaly detection (TSAD) is an important data mining task with numerous applications in the IoT era. In recent years, a large number of deep neural network-based methods have been proposed, demonstrating significantly better performance than conventional methods on addressing challenging TSAD problems in a variety of areas. Nevertheless, these deep TSAD methods typically rely on a clean training dataset that is not polluted by anomalies to learn the "normal profile" of the underlying dynamics. This requirement is nontrivial since a clean dataset can hardly be provided in practice. Moreover, without the awareness of their robustness, blindly applying deep TSAD methods with potentially contaminated training data can possibly incur significant performance degradation in the detection phase. In this work, to tackle this important challenge, we firstly investigate the robustness of commonly used deep TSAD methods with contaminated training data which provides a guideline for applying these methods when the provided training data are not guaranteed to be anomaly-free. Furthermore, we propose a model-agnostic method which can effectively improve the robustness of learning mainstream deep TSAD models with potentially contaminated data. Experiment results show that our method can consistently prevent or mitigate performance degradation of mainstream deep TSAD models on widely used benchmark datasets.

EGSDE: Unpaired Image-to-Image Translation via Energy-Guided Stochastic Differential Equations

Score-based diffusion generative models (SDGMs) have achieved the SOTA FID results in unpaired image-to-image translation (I2I). However, we notice that existing methods totally ignore the training data in the source domain, leading to sub-optimal solutions for unpaired I2I. To this end, we propose energy-guided stochastic differential equations (EGSDE) that employs an energy function pretrained on both the source and target domains to guide the inference process of a pretrained SDE for realistic and faithful unpaired I2I. Building upon two feature extractors, we carefully design the energy function such that it encourages the transferred image to preserve the domain-independent features and discard domainspecific ones. Further, we provide an alternative explanation of the EGSDE as a product of experts, where each of the three experts (corresponding to the SDE and two feature extractors) solely contributes to faithfulness or realism. Empirically, we compare EGSDE to a large family of baselines on three widely-adopted unpaired I2I tasks under four metrics. EGSDE not only consistently outperforms existing SDGMs-based methods in almost all settings but also achieves the SOTA realism results (e.g., FID of 65.82 in Cat to Dog and FID of 59.75 in Wild to Dog on AFHQ) without harming the faithful performance.

CoSCL: Cooperation of Small Continual Learners is Stronger than a Big One

Continual learning requires incremental compatibility with a sequence of tasks. However, the design of model architecture remains an open question: In general, learning all tasks with a shared set of parameters suffers from severe interference between tasks; while learning each task with a dedicated parameter subspace is limited by scalability. In this work, we theoretically analyze the generalization errors for learning plasticity and memory stability in continual learning, which can be uniformly upper-bounded by (1) discrepancy between task distributions, (2) flatness of loss landscape and (3) cover of parameter space. Then, inspired by the robust biological learning system that processes sequential experiences with multiple parallel compartments, we propose Cooperation of Small Continual Learners (CoSCL) as a general strategy for continual learning. Specifically, we present an architecture with a fixed number of narrower sub-networks to learn all incremental tasks in parallel, which can naturally reduce the two errors through improving the three components of the upper bound. To strengthen this advantage, we encourage to cooperate these sub-networks by penalizing the difference of predictions made by their feature representations. With a fixed parameter budget, CoSCL can improve a variety of representative continual learning approaches by a large margin (e.g., up to 10.64% on CIFAR-100-SC, 9.33% on CIFAR-100-RS, 11.45% on CUB-200-2011 and 6.72% on Tiny-ImageNet) and achieve the new state-of-the-art performance.

DGPO: Discovering Multiple Strategies with Diversity-Guided Policy Optimization

Recent algorithms designed for reinforcement learning tasks focus on finding a single optimal solution. However, in many practical applications, it is important to develop reasonable agents with diverse strategies. In this paper, we propose Diversity-Guided Policy Optimization (DGPO), an on-policy framework for discovering multiple strategies for the same task. Our algorithm uses diversity objectives to guide a latent code conditioned policy to learn a set of diverse strategies in a single training procedure. Specifically, we formalize our algorithm as the combination of a diversity-constrained optimization problem and an extrinsic-reward constrained optimization problem. And we solve the constrained optimization as a probabilistic inference task and use policy iteration to maximize the derived lower bound. Experimental results show that our method efficiently finds diverse strategies in a wide variety of reinforcement learning tasks. We further show that DGPO achieves a higher diversity score and has similar sample complexity and performance compared to other baselines.

Maximum Likelihood Training for Score-Based Diffusion ODEs by High-Order Denoising Score Matching

Score-based generative models have excellent performance in terms of generation quality and likelihood. They model the data distribution by matching a parameterized score network with first-order data score functions. The score network can be used to define an ODE ("score-based diffusion ODE") for exact likelihood evaluation. However, the relationship between the likelihood of the ODE and the score matching objective is unclear. In this work, we prove that matching the first-order score is not sufficient to maximize the likelihood of the ODE, by showing a gap between the maximum likelihood and score matching objectives. To fill up this gap, we show that the negative likelihood of the ODE can be bounded by controlling the first, second, and third-order score matching errors; and we further present a novel high-order denoising score matching method to enable maximum likelihood training of score-based diffusion ODEs. Our algorithm guarantees that the higher-order matching error is bounded by the training error and the lower-order errors. We empirically observe that by high-order score matching, score-based diffusion ODEs achieve better likelihood on both synthetic data and CIFAR-10, while retaining the high generation quality.