Xi Chen

Xi Chen, Tianyu Shi, Qingpeng Zhao et al.

Recent advances in deep reinforcement learning (RL) have demonstrated complex decision-making capabilities in simulation environments such as Arcade Learning Environment, MuJoCo, and ViZDoom. However, they are hardly extensible to more complicated problems, mainly due to the lack of complexity and variations in the environments they are trained and tested on. Furthermore, they are not extensible to an open-world environment to facilitate long-term exploration research. To learn realistic task-solving capabilities, we need to develop an environment with greater diversity and complexity. We developed WILD-SCAV, a powerful and extensible environment based on a 3D open-world FPS (First-Person Shooter) game to bridge the gap. It provides realistic 3D environments of variable complexity, various tasks, and multiple modes of interaction, where agents can learn to perceive 3D environments, navigate and plan, compete and cooperate in a human-like manner. WILD-SCAV also supports different complexities, such as configurable maps with different terrains, building structures and distributions, and multi-agent settings with cooperative and competitive tasks. The experimental results on configurable complexity, multi-tasking, and multi-agent scenarios demonstrate the effectiveness of WILD-SCAV in benchmarking various RL algorithms, as well as it is potential to give rise to intelligent agents with generalized task-solving abilities. The link to our open-sourced code can be found here https://github.com/inspirai/wilderness-scavenger.

Zifan Wang, Nan Ding, Tomer Levinboim et al. · deepmind

Recent research in robust optimization has shown an overfitting-like phenomenon in which models trained against adversarial attacks exhibit higher robustness on the training set compared to the test set. Although previous work provided theoretical explanations for this phenomenon using a robust PAC-Bayesian bound over the adversarial test error, related algorithmic derivations are at best only loosely connected to this bound, which implies that there is still a gap between their empirical success and our understanding of adversarial robustness theory. To close this gap, in this paper we consider a different form of the robust PAC-Bayesian bound and directly minimize it with respect to the model posterior. The derivation of the optimal solution connects PAC-Bayesian learning to the geometry of the robust loss surface through a Trace of Hessian (TrH) regularizer that measures the surface flatness. In practice, we restrict the TrH regularizer to the top layer only, which results in an analytical solution to the bound whose computational cost does not depend on the network depth. Finally, we evaluate our TrH regularization approach over CIFAR-10/100 and ImageNet using Vision Transformers (ViT) and compare against baseline adversarial robustness algorithms. Experimental results show that TrH regularization leads to improved ViT robustness that either matches or surpasses previous state-of-the-art approaches while at the same time requires less memory and computational cost.

Xi Chen, Christos Papadimitriou, Binghui Peng

Continual learning, or lifelong learning, is a formidable current challenge to machine learning. It requires the learner to solve a sequence of $k$ different learning tasks, one after the other, while retaining its aptitude for earlier tasks; the continual learner should scale better than the obvious solution of developing and maintaining a separate learner for each of the $k$ tasks. We embark on a complexity-theoretic study of continual learning in the PAC framework. We make novel uses of communication complexity to establish that any continual learner, even an improper one, needs memory that grows linearly with $k$, strongly suggesting that the problem is intractable. When logarithmically many passes over the learning tasks are allowed, we provide an algorithm based on multiplicative weights update whose memory requirement scales well; we also establish that improper learning is necessary for such performance. We conjecture that these results may lead to new promising approaches to continual learning.

Yongcheng Ding, José D. Martín-Guerrero, Yolanda Vives-Gilabert et al.

Active Learning (AL) is a family of machine learning (ML) algorithms that predates the current era of artificial intelligence. Unlike traditional approaches that require labeled samples for training, AL iteratively selects unlabeled samples to be annotated by an expert. This protocol aims to prioritize the most informative samples, leading to improved model performance compared to training with all labeled samples. In recent years, AL has gained increasing attention, particularly in the field of physics. This paper presents a comprehensive and accessible introduction to the theory of AL reviewing the latest advancements across various domains. Additionally, we explore the potential integration of AL with quantum ML, envisioning a synergistic fusion of these two fields rather than viewing AL as a mere extension of classical ML into the quantum realm.

Xi Chen, Cassandra Marcussen

We give an algorithm for testing uniformity of distributions supported on hypergrids $[m_1] \times \cdots \times [m_n]$, which makes $\smash{\widetilde{O}(\text{poly}(m)\sqrt{n}/ε^2)}$ many queries to a subcube conditional sampling oracle with $m=\max_i m_i$. When $m$ is a constant, our algorithm is nearly optimal and strengthens the algorithm of [CCK+21] which has the same query complexity but works for hypercubes $\{\pm 1\}^n$ only. A key technical contribution behind the analysis of our algorithm is a proof of a robust version of Pisier's inequality for functions over hypergrids using Fourier analysis.

Yi Li, Songqi Wang, Yaping Zhao et al.

The rapid advancement of artificial intelligence (AI) has been marked by the large language models exhibiting human-like intelligence. However, these models also present unprecedented challenges to energy consumption and environmental sustainability. One promising solution is to revisit analogue computing, a technique that predates digital computing and exploits emerging analogue electronic devices, such as resistive memory, which features in-memory computing, high scalability, and nonvolatility. However, analogue computing still faces the same challenges as before: programming nonidealities and expensive programming due to the underlying devices physics. Here, we report a universal solution, software-hardware co-design using structural plasticity-inspired edge pruning to optimize the topology of a randomly weighted analogue resistive memory neural network. Software-wise, the topology of a randomly weighted neural network is optimized by pruning connections rather than precisely tuning resistive memory weights. Hardware-wise, we reveal the physical origin of the programming stochasticity using transmission electron microscopy, which is leveraged for large-scale and low-cost implementation of an overparameterized random neural network containing high-performance sub-networks. We implemented the co-design on a 40nm 256K resistive memory macro, observing 17.3% and 19.9% accuracy improvements in image and audio classification on FashionMNIST and Spoken digits datasets, as well as 9.8% (2%) improvement in PR (ROC) in image segmentation on DRIVE datasets, respectively. This is accompanied by 82.1%, 51.2%, and 99.8% improvement in energy efficiency thanks to analogue in-memory computing. By embracing the intrinsic stochasticity and in-memory computing, this work may solve the biggest obstacle of analogue computing systems and thus unleash their immense potential for next-generation AI hardware.

Xi Chen, Siwei Mai, Konstantinos Michmizos

A vast majority of spiking neural networks (SNNs) are trained based on inductive biases that are not necessarily a good fit for several critical tasks that require low-latency and power efficiency. Inferring brain behavior based on the associated electroenchephalography (EEG) signals is an example of how networks training and inference efficiency can be heavily impacted by learning spatio-temporal dependencies. Up to now, SNNs rely solely on general inductive biases to model the dynamic relations between different data streams. Here, we propose a graph spiking neural network architecture for multi-channel EEG classification (EEGSN) that learns the dynamic relational information present in the distributed EEG sensors. Our method reduced the inference computational complexity by $\times 20$ compared to the state-of-the-art SNNs, while achieved comparable accuracy on motor execution classification tasks. Overall, our work provides a framework for interpretable and efficient training of graph spiking networks that are suitable for low-latency and low-power real-time applications.

Rudrajit Das, Xi Chen, Bertram Ieong et al.

It is well known that selecting samples with large losses/gradients can significantly reduce the number of training steps. However, the selection overhead is often too high to yield any meaningful gains in terms of overall training time. In this work, we focus on the greedy approach of selecting samples with large \textit{approximate losses} instead of exact losses in order to reduce the selection overhead. For smooth convex losses, we show that such a greedy strategy can converge to a constant factor of the minimum value of the average loss in fewer iterations than the standard approach of random selection. We also theoretically quantify the effect of the approximation level. We then develop SIFT which uses early exiting to obtain approximate losses with an intermediate layer's representations for sample selection. We evaluate SIFT on the task of training a 110M parameter 12 layer BERT base model, and show significant gains (in terms of training hours and number of backpropagation steps) without any optimized implementation over vanilla training. For e.g., to reach 64% validation accuracy, SIFT with exit at the first layer takes ~ 43 hours compared to ~ 57 hours of vanilla training.

Yongcheng Ding, Xi Chen, Rafael Magdalena-Benedicto et al.

The exotic nature of quantum mechanics makes machine learning (ML) be different in the quantum realm compared to classical applications. ML can be used for knowledge discovery using information continuously extracted from a quantum system in a broad range of tasks. The model receives streaming quantum information for learning and decision-making, resulting in instant feedback on the quantum system. As a stream learning approach, we present a deep reinforcement learning on streaming data from a continuously measured qubit at the presence of detuning, dephasing, and relaxation. We also investigate how the agent adapts to another quantum noise pattern by transfer learning. Stream learning provides a better understanding of closed-loop quantum control, which may pave the way for advanced quantum technologies.

Xi Chen, Quanquan Liu, Yining Wang

In this paper, we study the learning problem in contextual search, which is motivated by applications such as first-price auction, personalized medicine experiments, and feature-based pricing experiments. In particular, for a sequence of arriving context vectors, with each context associated with an underlying value, the decision-maker either makes a query at a certain point or skips the context. The decision-maker will only observe the binary feedback on the relationship between the query point and the value associated with the context. We study a PAC learning setting, where the goal is to learn the underlying mean value function in context with a minimum number of queries. To address this challenge, we propose a tri-section search approach combined with a margin-based active learning method. We show that the algorithm only needs to make $O(1/\varepsilon^2)$ queries to achieve an $ε$-estimation accuracy. This sample complexity significantly reduces the required sample complexity in the passive setting, at least $Ω(1/\varepsilon^4)$.

Zifeng Wang, Shao-Lun Huang, Ercan E. Kuruoglu et al.

Understanding the source of the superior generalization ability of NNs remains one of the most important problems in ML research. There have been a series of theoretical works trying to derive non-vacuous bounds for NNs. Recently, the compression of information stored in weights (IIW) is proved to play a key role in NNs generalization based on the PAC-Bayes theorem. However, no solution of IIW has ever been provided, which builds a barrier for further investigation of the IIW's property and its potential in practical deep learning. In this paper, we propose an algorithm for the efficient approximation of IIW. Then, we build an IIW-based information bottleneck on the trade-off between accuracy and information complexity of NNs, namely PIB. From PIB, we can empirically identify the fitting to compressing phase transition during NNs' training and the concrete connection between the IIW compression and the generalization. Besides, we verify that IIW is able to explain NNs in broad cases, e.g., varying batch sizes, over-parameterization, and noisy labels. Moreover, we propose an MCMC-based algorithm to sample from the optimal weight posterior characterized by PIB, which fulfills the potential of IIW in enhancing NNs in practice.

Yongcheng Ding, José D. Martín-Guerrero, Yujing Song et al.

Optimal design for model training is a critical topic in machine learning. Active Learning aims at obtaining improved models by querying samples with maximum uncertainty according to the estimation model for artificially labeling; this has the additional advantage of achieving successful performances with a reduced number of labeled samples. We analyze its capability as an assistant for the design of experiments, extracting maximum information for learning with the minimal cost in fidelity loss, or reducing total operation costs of labeling in the laboratory. We present two typical applications as quantum information retrieval in qutrits and phase boundary prediction in many-body physics. For an equivalent multinomial classification problem, we achieve the correct rate of 99% with less than 2% samples labeled. We reckon that active-learning-inspired physics experiments will remarkably save budget without loss of accuracy.

Nan Ding, Xi Chen, Tomer Levinboim et al.

Despite recent advances in its theoretical understanding, there still remains a significant gap in the ability of existing PAC-Bayesian theories on meta-learning to explain performance improvements in the few-shot learning setting, where the number of training examples in the target tasks is severely limited. This gap originates from an assumption in the existing theories which supposes that the number of training examples in the observed tasks and the number of training examples in the target tasks follow the same distribution, an assumption that rarely holds in practice. By relaxing this assumption, we develop two PAC-Bayesian bounds tailored for the few-shot learning setting and show that two existing meta-learning algorithms (MAML and Reptile) can be derived from our bounds, thereby bridging the gap between practice and PAC-Bayesian theories. Furthermore, we derive a new computationally-efficient PACMAML algorithm, and show it outperforms existing meta-learning algorithms on several few-shot benchmark datasets.

Ali Ghadirzadeh, Xi Chen, Petra Poklukar et al.

Reinforcement learning methods can achieve significant performance but require a large amount of training data collected on the same robotic platform. A policy trained with expensive data is rendered useless after making even a minor change to the robot hardware. In this paper, we address the challenging problem of adapting a policy, trained to perform a task, to a novel robotic hardware platform given only few demonstrations of robot motion trajectories on the target robot. We formulate it as a few-shot meta-learning problem where the goal is to find a meta-model that captures the common structure shared across different robotic platforms such that data-efficient adaptation can be performed. We achieve such adaptation by introducing a learning framework consisting of a probabilistic gradient-based meta-learning algorithm that models the uncertainty arising from the few-shot setting with a low-dimensional latent variable. We experimentally evaluate our framework on a simulated reaching and a real-robot picking task using 400 simulated robots generated by varying the physical parameters of an existing set of robotic platforms. Our results show that the proposed method can successfully adapt a trained policy to different robotic platforms with novel physical parameters and the superiority of our meta-learning algorithm compared to state-of-the-art methods for the introduced few-shot policy adaptation problem.

Xi Chen, Zehua Lai, He Li et al.

This paper investigates the problem of online statistical inference of model parameters in stochastic optimization problems via the Kiefer-Wolfowitz algorithm with random search directions. We first present the asymptotic distribution for the Polyak-Ruppert-averaging type Kiefer-Wolfowitz (AKW) estimators, whose asymptotic covariance matrices depend on the distribution of search directions and the function-value query complexity. The distributional result reflects the trade-off between statistical efficiency and function query complexity. We further analyze the choice of random search directions to minimize certain summary statistics of the asymptotic covariance matrix. Based on the asymptotic distribution, we conduct online statistical inference by providing two construction procedures of valid confidence intervals.

S. Hu, X. Chen, W. Ni et al.

Distributed machine learning (DML) techniques, such as federated learning, partitioned learning, and distributed reinforcement learning, have been increasingly applied to wireless communications. This is due to improved capabilities of terminal devices, explosively growing data volume, congestion in the radio interfaces, and increasing concern of data privacy. The unique features of wireless systems, such as large scale, geographically dispersed deployment, user mobility, and massive amount of data, give rise to new challenges in the design of DML techniques. There is a clear gap in the existing literature in that the DML techniques are yet to be systematically reviewed for their applicability to wireless systems. This survey bridges the gap by providing a contemporary and comprehensive survey of DML techniques with a focus on wireless networks. Specifically, we review the latest applications of DML in power control, spectrum management, user association, and edge cloud computing. The optimality, scalability, convergence rate, computation cost, and communication overhead of DML are analyzed. We also discuss the potential adversarial attacks faced by DML applications, and describe state-of-the-art countermeasures to preserve privacy and security. Last but not least, we point out a number of key issues yet to be addressed, and collate potentially interesting and challenging topics for future research.

Xi Chen, Rajesh Jayaram, Amit Levi et al.

We study the problems of learning and testing junta distributions on $\{-1,1\}^n$ with respect to the uniform distribution, where a distribution $p$ is a $k$-junta if its probability mass function $p(x)$ depends on a subset of at most $k$ variables. The main contribution is an algorithm for finding relevant coordinates in a $k$-junta distribution with subcube conditioning [BC18, CCKLW20]. We give two applications: 1. An algorithm for learning $k$-junta distributions with $\tilde{O}(k/ε^2) \log n + O(2^k/ε^2)$ subcube conditioning queries, and 2. An algorithm for testing $k$-junta distributions with $\tilde{O}((k + \sqrt{n})/ε^2)$ subcube conditioning queries. All our algorithms are optimal up to poly-logarithmic factors. Our results show that subcube conditioning, as a natural model for accessing high-dimensional distributions, enables significant savings in learning and testing junta distributions compared to the standard sampling model. This addresses an open question posed by Aliakbarpour, Blais, and Rubinfeld [ABR17].

Clément L. Canonne, Xi Chen, Gautam Kamath et al.

We give a nearly-optimal algorithm for testing uniformity of distributions supported on $\{-1,1\}^n$, which makes $\tilde O (\sqrt{n}/\varepsilon^2)$ queries to a subcube conditional sampling oracle (Bhattacharyya and Chakraborty (2018)). The key technical component is a natural notion of random restriction for distributions on $\{-1,1\}^n$, and a quantitative analysis of how such a restriction affects the mean vector of the distribution. Along the way, we consider the problem of mean testing with independent samples and provide a nearly-optimal algorithm.

Xi Chen, Ali Ghadirzadeh, Mårten Björkman et al.

Deep reinforcement learning (RL) has enabled training action-selection policies, end-to-end, by learning a function which maps image pixels to action outputs. However, it's application to visuomotor robotic policy training has been limited because of the challenge of large-scale data collection when working with physical hardware. A suitable visuomotor policy should perform well not just for the task-setup it has been trained for, but also for all varieties of the task, including novel objects at different viewpoints surrounded by task-irrelevant objects. However, it is impractical for a robotic setup to sufficiently collect interactive samples in a RL framework to generalize well to novel aspects of a task. In this work, we demonstrate that by using adversarial training for domain transfer, it is possible to train visuomotor policies based on RL frameworks, and then transfer the acquired policy to other novel task domains. We propose to leverage the deep RL capabilities to learn complex visuomotor skills for uncomplicated task setups, and then exploit transfer learning to generalize to new task domains provided only still images of the task in the target domain. We evaluate our method on two real robotic tasks, picking and pouring, and compare it to a number of prior works, demonstrating its superiority.

Yang Liu, Yunan Luo, Yuanyi Zhong et al.

Recent advances in deep reinforcement learning algorithms have shown great potential and success for solving many challenging real-world problems, including Go game and robotic applications. Usually, these algorithms need a carefully designed reward function to guide training in each time step. However, in real world, it is non-trivial to design such a reward function, and the only signal available is usually obtained at the end of a trajectory, also known as the episodic reward or return. In this work, we introduce a new algorithm for temporal credit assignment, which learns to decompose the episodic return back to each time-step in the trajectory using deep neural networks. With this learned reward signal, the learning efficiency can be substantially improved for episodic reinforcement learning. In particular, we find that expressive language models such as the Transformer can be adopted for learning the importance and the dependency of states in the trajectory, therefore providing high-quality and interpretable learned reward signals. We have performed extensive experiments on a set of MuJoCo continuous locomotive control tasks with only episodic returns and demonstrated the effectiveness of our algorithm.

Feng Hu, Lucas Lamata, Mikel Sanz et al.

As the building block in symmetric cryptography, designing Boolean functions satisfying multiple properties is an important problem in sequence ciphers, block ciphers, and hash functions. However, the search of $n$-variable Boolean functions fulfilling global cryptographic constraints is computationally hard due to the super-exponential size $\mathcal{O}(2^{2^n})$ of the space. Here, we introduce a codification of the cryptographically relevant constraints in the ground state of an Ising Hamiltonian, allowing us to naturally encode it in a quantum annealer, which seems to provide a quantum speedup. Additionally, we benchmark small $n$ cases in a D-Wave machine, showing its capacity of devising bent functions, the most relevant set of cryptographic Boolean functions. We have complemented it with local search and chain repair to improve the D-Wave quantum annealer performance related to the low connectivity. This work shows how to codify super-exponential cryptographic problems into quantum annealers and paves the way for reaching quantum supremacy with an adequately designed chip.

Xi Chen, Zhihong Deng, Gehui Shen et al.

This paper proposes a novel framework for recurrent neural networks (RNNs) inspired by the human memory models in the field of cognitive neuroscience to enhance information processing and transmission between adjacent RNNs' units. The proposed framework for RNNs consists of three stages that is working memory, forget, and long-term store. The first stage includes taking input data into sensory memory and transferring it to working memory for preliminary treatment. And the second stage mainly focuses on proactively forgetting the secondary information rather than the primary in the working memory. And finally, we get the long-term store normally using some kind of RNN's unit. Our framework, which is generalized and simple, is evaluated on 6 datasets which fall into 3 different tasks, corresponding to text classification, image classification and language modelling. Experiments reveal that our framework can obviously improve the performance of traditional recurrent neural networks. And exploratory task shows the ability of our framework of correctly forgetting the secondary information.

Xi Chen, Ali Ghadirzadeh, John Folkesson et al.

Mobile robot navigation in complex and dynamic environments is a challenging but important problem. Reinforcement learning approaches fail to solve these tasks efficiently due to reward sparsities, temporal complexities and high-dimensionality of sensorimotor spaces which are inherent in such problems. We present a novel approach to train action policies to acquire navigation skills for wheel-legged robots using deep reinforcement learning. The policy maps height-map image observations to motor commands to navigate to a target position while avoiding obstacles. We propose to acquire the multifaceted navigation skill by learning and exploiting a number of manageable navigation behaviors. We also introduce a domain randomization technique to improve the versatility of the training samples. We demonstrate experimentally a significant improvement in terms of data-efficiency, success rate, robustness against irrelevant sensory data, and also the quality of the maneuver skills.

Bradly C. Stadie, Ge Yang, Rein Houthooft et al.

We consider the problem of exploration in meta reinforcement learning. Two new meta reinforcement learning algorithms are suggested: E-MAML and E-$\text{RL}^2$. Results are presented on a novel environment we call `Krazy World' and a set of maze environments. We show E-MAML and E-$\text{RL}^2$ deliver better performance on tasks where exploration is important.

John Schulman, Xi Chen, Pieter Abbeel

Two of the leading approaches for model-free reinforcement learning are policy gradient methods and $Q$-learning methods. $Q$-learning methods can be effective and sample-efficient when they work, however, it is not well-understood why they work, since empirically, the $Q$-values they estimate are very inaccurate. A partial explanation may be that $Q$-learning methods are secretly implementing policy gradient updates: we show that there is a precise equivalence between $Q$-learning and policy gradient methods in the setting of entropy-regularized reinforcement learning, that "soft" (entropy-regularized) $Q$-learning is exactly equivalent to a policy gradient method. We also point out a connection between $Q$-learning methods and natural policy gradient methods. Experimentally, we explore the entropy-regularized versions of $Q$-learning and policy gradients, and we find them to perform as well as (or slightly better than) the standard variants on the Atari benchmark. We also show that the equivalence holds in practical settings by constructing a $Q$-learning method that closely matches the learning dynamics of A3C without using a target network or $ε$-greedy exploration schedule.

Haoran Tang, Rein Houthooft, Davis Foote et al.

Count-based exploration algorithms are known to perform near-optimally when used in conjunction with tabular reinforcement learning (RL) methods for solving small discrete Markov decision processes (MDPs). It is generally thought that count-based methods cannot be applied in high-dimensional state spaces, since most states will only occur once. Recent deep RL exploration strategies are able to deal with high-dimensional continuous state spaces through complex heuristics, often relying on optimism in the face of uncertainty or intrinsic motivation. In this work, we describe a surprising finding: a simple generalization of the classic count-based approach can reach near state-of-the-art performance on various high-dimensional and/or continuous deep RL benchmarks. States are mapped to hash codes, which allows to count their occurrences with a hash table. These counts are then used to compute a reward bonus according to the classic count-based exploration theory. We find that simple hash functions can achieve surprisingly good results on many challenging tasks. Furthermore, we show that a domain-dependent learned hash code may further improve these results. Detailed analysis reveals important aspects of a good hash function: 1) having appropriate granularity and 2) encoding information relevant to solving the MDP. This exploration strategy achieves near state-of-the-art performance on both continuous control tasks and Atari 2600 games, hence providing a simple yet powerful baseline for solving MDPs that require considerable exploration.

Yan Duan, John Schulman, Xi Chen et al.

Deep reinforcement learning (deep RL) has been successful in learning sophisticated behaviors automatically; however, the learning process requires a huge number of trials. In contrast, animals can learn new tasks in just a few trials, benefiting from their prior knowledge about the world. This paper seeks to bridge this gap. Rather than designing a "fast" reinforcement learning algorithm, we propose to represent it as a recurrent neural network (RNN) and learn it from data. In our proposed method, RL$^2$, the algorithm is encoded in the weights of the RNN, which are learned slowly through a general-purpose ("slow") RL algorithm. The RNN receives all information a typical RL algorithm would receive, including observations, actions, rewards, and termination flags; and it retains its state across episodes in a given Markov Decision Process (MDP). The activations of the RNN store the state of the "fast" RL algorithm on the current (previously unseen) MDP. We evaluate RL$^2$ experimentally on both small-scale and large-scale problems. On the small-scale side, we train it to solve randomly generated multi-arm bandit problems and finite MDPs. After RL$^2$ is trained, its performance on new MDPs is close to human-designed algorithms with optimality guarantees. On the large-scale side, we test RL$^2$ on a vision-based navigation task and show that it scales up to high-dimensional problems.