arXiv subjects

Zhilong Zhang

Publications and source records attributed to Zhilong Zhang.

35 records · Page 2Linked to original sources

arXiv2024

Stable Continual Reinforcement Learning via Diffusion-based Trajectory Replay

Given the inherent non-stationarity prevalent in real-world applications, continual Reinforcement Learning (RL) aims to equip the agent with the capability to address a series of sequentially presented decision-making tasks. Within this problem setting, a pivotal challenge revolves around \textit{catastrophic forgetting} issue, wherein the agent is prone to effortlessly erode the decisional knowledge associated with past encountered tasks when learning the new one. In recent progresses, the \textit{generative replay} methods have showcased substantial potential by employing generative models to replay data distribution of past tasks. Compared to storing the data from past tasks directly, this category of methods circumvents the growing storage overhead and possible data privacy concerns. However, constrained by the expressive capacity of generative models, existing \textit{generative replay} methods face challenges in faithfully reconstructing the data distribution of past tasks, particularly in scenarios with a myriad of tasks or high-dimensional data. Inspired by the success of diffusion models in various generative tasks, this paper introduces a novel continual RL algorithm DISTR (Diffusion-based Trajectory Replay) that employs a diffusion model to memorize the high-return trajectory distribution of each encountered task and wakeups these distributions during the policy learning on new tasks. Besides, considering the impracticality of replaying all past data each time, a prioritization mechanism is proposed to prioritize the trajectory replay of pivotal tasks in our method. Empirical experiments on the popular continual RL benchmark \texttt{Continual World} demonstrate that our proposed method obtains a favorable balance between \textit{stability} and \textit{plasticity}, surpassing various existing continual RL baselines in average success rate.

cs.LG↗

arXiv2024

WHALE: Towards Generalizable and Scalable World Models for Embodied Decision-making

World models play a crucial role in decision-making within embodied environments, enabling cost-free explorations that would otherwise be expensive in the real world. To facilitate effective decision-making, world models must be equipped with strong generalizability to support faithful imagination in out-of-distribution (OOD) regions and provide reliable uncertainty estimation to assess the credibility of the simulated experiences, both of which present significant challenges for prior scalable approaches. This paper introduces WHALE, a framework for learning generalizable world models, consisting of two key techniques: behavior-conditioning and retracing-rollout. Behavior-conditioning addresses the policy distribution shift, one of the primary sources of the world model generalization error, while retracing-rollout enables efficient uncertainty estimation without the necessity of model ensembles. These techniques are universal and can be combined with any neural network architecture for world model learning. Incorporating these two techniques, we present Whale-ST, a scalable spatial-temporal transformer-based world model with enhanced generalizability. We demonstrate the superiority of Whale-ST in simulation tasks by evaluating both value estimation accuracy and video generation fidelity. Additionally, we examine the effectiveness of our uncertainty estimation technique, which enhances model-based policy optimization in fully offline scenarios. Furthermore, we propose Whale-X, a 414M parameter world model trained on 970K trajectories from Open X-Embodiment datasets. We show that Whale-X exhibits promising scalability and strong generalizability in real-world manipulation scenarios using minimal demonstrations.

cs.LG↗

arXiv2024

Provably and Practically Efficient Adversarial Imitation Learning with General Function Approximation

As a prominent category of imitation learning methods, adversarial imitation learning (AIL) has garnered significant practical success powered by neural network approximation. However, existing theoretical studies on AIL are primarily limited to simplified scenarios such as tabular and linear function approximation and involve complex algorithmic designs that hinder practical implementation, highlighting a gap between theory and practice. In this paper, we explore the theoretical underpinnings of online AIL with general function approximation. We introduce a new method called optimization-based AIL (OPT-AIL), which centers on performing online optimization for reward functions and optimism-regularized Bellman error minimization for Q-value functions. Theoretically, we prove that OPT-AIL achieves polynomial expert sample complexity and interaction complexity for learning near-expert policies. To our best knowledge, OPT-AIL is the first provably efficient AIL method with general function approximation. Practically, OPT-AIL only requires the approximate optimization of two objectives, thereby facilitating practical implementation. Empirical studies demonstrate that OPT-AIL outperforms previous state-of-the-art deep AIL methods in several challenging tasks.

cs.LG↗

arXiv2024

Energy-Guided Diffusion Sampling for Offline-to-Online Reinforcement Learning

Combining offline and online reinforcement learning (RL) techniques is indeed crucial for achieving efficient and safe learning where data acquisition is expensive. Existing methods replay offline data directly in the online phase, resulting in a significant challenge of data distribution shift and subsequently causing inefficiency in online fine-tuning. To address this issue, we introduce an innovative approach, \textbf{E}nergy-guided \textbf{DI}ffusion \textbf{S}ampling (EDIS), which utilizes a diffusion model to extract prior knowledge from the offline dataset and employs energy functions to distill this knowledge for enhanced data generation in the online phase. The theoretical analysis demonstrates that EDIS exhibits reduced suboptimality compared to solely utilizing online data or directly reusing offline data. EDIS is a plug-in approach and can be combined with existing methods in offline-to-online RL setting. By implementing EDIS to off-the-shelf methods Cal-QL and IQL, we observe a notable 20% average improvement in empirical performance on MuJoCo, AntMaze, and Adroit environments. Code is available at \url{https://github.com/liuxhym/EDIS}.

cs.LG↗

arXiv2024

Consistency Flow Matching: Defining Straight Flows with Velocity Consistency

Flow matching (FM) is a general framework for defining probability paths via Ordinary Differential Equations (ODEs) to transform between noise and data samples. Recent approaches attempt to straighten these flow trajectories to generate high-quality samples with fewer function evaluations, typically through iterative rectification methods or optimal transport solutions. In this paper, we introduce Consistency Flow Matching (Consistency-FM), a novel FM method that explicitly enforces self-consistency in the velocity field. Consistency-FM directly defines straight flows starting from different times to the same endpoint, imposing constraints on their velocity values. Additionally, we propose a multi-segment training approach for Consistency-FM to enhance expressiveness, achieving a better trade-off between sampling quality and speed. Preliminary experiments demonstrate that our Consistency-FM significantly improves training efficiency by converging 4.4x faster than consistency models and 1.7x faster than rectified flow models while achieving better generation quality. Our code is available at: https://github.com/YangLing0818/consistency_flow_matching

cs.CV↗

arXiv2024

Contextualized Diffusion Models for Text-Guided Image and Video Generation

Conditional diffusion models have exhibited superior performance in high-fidelity text-guided visual generation and editing. Nevertheless, prevailing text-guided visual diffusion models primarily focus on incorporating text-visual relationships exclusively into the reverse process, often disregarding their relevance in the forward process. This inconsistency between forward and reverse processes may limit the precise conveyance of textual semantics in visual synthesis results. To address this issue, we propose a novel and general contextualized diffusion model (ContextDiff) by incorporating the cross-modal context encompassing interactions and alignments between text condition and visual sample into forward and reverse processes. We propagate this context to all timesteps in the two processes to adapt their trajectories, thereby facilitating cross-modal conditional modeling. We generalize our contextualized diffusion to both DDPMs and DDIMs with theoretical derivations, and demonstrate the effectiveness of our model in evaluations with two challenging tasks: text-to-image generation, and text-to-video editing. In each task, our ContextDiff achieves new state-of-the-art performance, significantly enhancing the semantic alignment between text condition and generated samples, as evidenced by quantitative and qualitative evaluations. Our code is available at https://github.com/YangLing0818/ContextDiff

cs.CV↗

arXiv2024

Any-step Dynamics Model Improves Future Predictions for Online and Offline Reinforcement Learning

Model-based methods in reinforcement learning offer a promising approach to enhance data efficiency by facilitating policy exploration within a dynamics model. However, accurately predicting sequential steps in the dynamics model remains a challenge due to the bootstrapping prediction, which attributes the next state to the prediction of the current state. This leads to accumulated errors during model roll-out. In this paper, we propose the Any-step Dynamics Model (ADM) to mitigate the compounding error by reducing bootstrapping prediction to direct prediction. ADM allows for the use of variable-length plans as inputs for predicting future states without frequent bootstrapping. We design two algorithms, ADMPO-ON and ADMPO-OFF, which apply ADM in online and offline model-based frameworks, respectively. In the online setting, ADMPO-ON demonstrates improved sample efficiency compared to previous state-of-the-art methods. In the offline setting, ADMPO-OFF not only demonstrates superior performance compared to recent state-of-the-art offline approaches but also offers better quantification of model uncertainty using only a single ADM.

cs.LG↗

arXiv2024

BWArea Model: Learning World Model, Inverse Dynamics, and Policy for Controllable Language Generation

Large language models (LLMs) have catalyzed a paradigm shift in natural language processing, yet their limited controllability poses a significant challenge for downstream applications. We aim to address this by drawing inspiration from the neural mechanisms of the human brain, specifically Broca's and Wernicke's areas, which are crucial for language generation and comprehension, respectively. In particular, Broca's area receives cognitive decision signals from Wernicke's area, treating the language generation as an intricate decision-making process, which differs from the fully auto-regressive language generation of existing LLMs. In a similar vein, our proposed system, the BWArea model, conceptualizes language generation as a decision-making task. This model has three components: a language world model, an inverse dynamics model, and a cognitive policy. Like Wernicke's area, the inverse dynamics model is designed to deduce the underlying cognitive intentions, or latent actions, behind each token. The BWArea model is amenable to both pre-training and fine-tuning like existing LLMs. With 30B clean pre-training tokens, we have trained a BWArea model, which achieves competitive performance with LLMs of equal size (1B parameters). Unlike fully auto-regressive LLMs, its pre-training performance does not degenerate if dirty data unintentionally appears. This shows the advantage of a decomposed structure of BWArea model in reducing efforts in laborious data selection and labeling. Finally, we reveal that the BWArea model offers enhanced controllability via fine-tuning the cognitive policy with downstream reward metrics, thereby facilitating alignment with greater simplicity. On 9 out of 10 tasks from two suites, TextWorld and BigBench Hard, our method shows superior performance to auto-regressive LLMs.

cs.CL↗

arXiv2024

Jointly Optimizing Terahertz based Sensing and Communications in Vehicular Networks: A Dynamic Graph Neural Network Approach

In this paper, the problem of vehicle service mode selection (sensing, communication, or both) and vehicle connections within terahertz (THz) enabled joint sensing and communications over vehicular networks is studied. The considered network consists of several service provider vehicles (SPVs) that can provide: 1) only sensing service, 2) only communication service, and 3) both services, sensing service request vehicles, and communication service request vehicles. Based on the vehicle network topology and their service accessibility, SPVs strategically select service request vehicles to provide sensing, communication, or both services. This problem is formulated as an optimization problem, aiming to maximize the number of successfully served vehicles by jointly determining the service mode of each SPV and its associated vehicles. To solve this problem, we propose a dynamic graph neural network (GNN) model that selects appropriate graph information aggregation functions according to the vehicle network topology, thus extracting more vehicle network information compared to traditional static GNNs that use fixed aggregation functions for different vehicle network topologies. Using the extracted vehicle network information, the service mode of each SPV and its served service request vehicles will be determined. Simulation results show that the proposed dynamic GNN based method can improve the number of successfully served vehicles by up to 17% and 28% compared to a GNN based algorithm with a fixed neural network model and a conventional optimization algorithm without using GNNs.

cs.NI↗

arXiv2024

Structure-Guided Adversarial Training of Diffusion Models

Diffusion models have demonstrated exceptional efficacy in various generative applications. While existing models focus on minimizing a weighted sum of denoising score matching losses for data distribution modeling, their training primarily emphasizes instance-level optimization, overlooking valuable structural information within each mini-batch, indicative of pair-wise relationships among samples. To address this limitation, we introduce Structure-guided Adversarial training of Diffusion Models (SADM). In this pioneering approach, we compel the model to learn manifold structures between samples in each training batch. To ensure the model captures authentic manifold structures in the data distribution, we advocate adversarial training of the diffusion generator against a novel structure discriminator in a minimax game, distinguishing real manifold structures from the generated ones. SADM substantially improves existing diffusion transformers (DiT) and outperforms existing methods in image generation and cross-domain fine-tuning tasks across 12 datasets, establishing a new state-of-the-art FID of 1.58 and 2.11 on ImageNet for class-conditional image generation at resolutions of 256x256 and 512x512, respectively.

cs.CV↗

arXiv2024

Importance-Aware Image Segmentation-based Semantic Communication for Autonomous Driving

This article studies the problem of image segmentation-based semantic communication in autonomous driving. In real traffic scenes, detecting the key objects (e.g., vehicles, pedestrians and obstacles) is more crucial than that of other objects to guarantee driving safety. Therefore, we propose a vehicular image segmentation-oriented semantic communication system, termed VIS-SemCom, where image segmentation features of important objects are transmitted to reduce transmission redundancy. First, to accurately extract image semantics, we develop a semantic codec based on Swin Transformer architecture, which expands the perceptual field thus improving the segmentation accuracy. Next, we propose a multi-scale semantic extraction scheme via assigning the number of Swin Transformer blocks for diverse resolution features, thus highlighting the important objects' accuracy. Furthermore, the importance-aware loss is invoked to emphasize the important objects, and an online hard sample mining (OHEM) strategy is proposed to handle small sample issues in the dataset. Experimental results demonstrate that the proposed VIS-SemCom can achieve a coding gain of nearly 6 dB with a 60% mean intersection over union (mIoU), reduce the transmitted data amount by up to 70% with a 60% mIoU, and improve the segmentation intersection over union (IoU) of important objects by 4%, compared to traditional transmission scheme.

cs.NI↗

arXiv2024

Improving Diffusion-Based Image Synthesis with Context Prediction

Diffusion models are a new class of generative models, and have dramatically promoted image generation with unprecedented quality and diversity. Existing diffusion models mainly try to reconstruct input image from a corrupted one with a pixel-wise or feature-wise constraint along spatial axes. However, such point-based reconstruction may fail to make each predicted pixel/feature fully preserve its neighborhood context, impairing diffusion-based image synthesis. As a powerful source of automatic supervisory signal, context has been well studied for learning representations. Inspired by this, we for the first time propose ConPreDiff to improve diffusion-based image synthesis with context prediction. We explicitly reinforce each point to predict its neighborhood context (i.e., multi-stride features/tokens/pixels) with a context decoder at the end of diffusion denoising blocks in training stage, and remove the decoder for inference. In this way, each point can better reconstruct itself by preserving its semantic connections with neighborhood context. This new paradigm of ConPreDiff can generalize to arbitrary discrete and continuous diffusion backbones without introducing extra parameters in sampling procedure. Extensive experiments are conducted on unconditional image generation, text-to-image generation and image inpainting tasks. Our ConPreDiff consistently outperforms previous methods and achieves a new SOTA text-to-image generation results on MS-COCO, with a zero-shot FID score of 6.21.

cs.CV↗

arXiv2023

Identifying Subgroups of ICU Patients Using End-to-End Multivariate Time-Series Clustering Algorithm Based on Real-World Vital Signs Data

This study employed the MIMIC-IV database as data source to investigate the use of dynamic, high-frequency, multivariate time-series vital signs data, including temperature, heart rate, mean blood pressure, respiratory rate, and SpO2, monitored first 8 hours data in the ICU stay. Various clustering algorithms were compared, and an end-to-end multivariate time series clustering system called Time2Feat, combined with K-Means, was chosen as the most effective method to cluster patients in the ICU. In clustering analysis, data of 8,080 patients admitted between 2008 and 2016 was used for model development and 2,038 patients admitted between 2017 and 2019 for model validation. By analyzing the differences in clinical mortality prognosis among different categories, varying risks of ICU mortality and hospital mortality were found between different subgroups. Furthermore, the study visualized the trajectory of vital signs changes. The findings of this study provide valuable insights into the potential use of multivariate time-series clustering systems in patient management and monitoring in the ICU setting.

cs.LG↗

arXiv2023

Graph Neural Networks for Joint Communication and Sensing Optimization in Vehicular Networks

In this paper, the problem of joint communication and sensing is studied in the context of terahertz (THz) vehicular networks. In the studied model, a set of service provider vehicles (SPVs) provide either communication service or sensing service to target vehicles, where it is essential to determine 1) the service mode (i.e., providing either communication or sensing service) for each SPV and 2) the subset of target vehicles that each SPV will serve. The problem is formulated as an optimization problem aiming to maximize the sum of the data rates of the communication target vehicles, while satisfying the sensing service requirements of the sensing target vehicles, by determining the service mode and the target vehicle association for each SPV. To solve this problem, a graph neural network (GNN) based algorithm with a heterogeneous graph representation is proposed. The proposed algorithm enables the central controller to extract each vehicle's graph information related to its location, connection, and communication interference. Using this extracted graph information, a joint service mode selection and target vehicle association strategy is then determined to adapt to the dynamic vehicle topology with various vehicle types (e.g., target vehicles and service provider vehicles). Simulation results show that the proposed GNN-based scheme can achieve 93.66% of the sum rate achieved by the optimal solution, and yield up to 3.16% and 31.86% improvements in sum rate, respectively, over a homogeneous GNN-based algorithm and a conventional optimization algorithm without using GNNs.

cs.NI↗

arXiv2022

Image Segmentation Semantic Communication over Internet of Vehicles

In this paper, the problem of semantic-based efficient image transmission is studied over the Internet of Vehicles (IoV). In the considered model, a vehicle shares massive amount of visual data perceived by its visual sensors to assist other vehicles in making driving decisions. However, it is hard to maintain a high reliable visual data transmission due to the limited spectrum resources. To tackle this problem, a semantic communication approach is introduced to reduce the transmission data amount while ensuring the semantic-level accuracy. Particularly, an image segmentation semantic communication (ISSC) system is proposed, which can extract the semantic features from the perceived images and transmit the features to the receiving vehicle that reconstructs the image segmentations. The ISSC system consists of an encoder and a decoder at the transmitter and the receiver, respectively. To accurately extract the image semantic features, the ISSC system encoder employs a Swin Transformer based multi-scale semantic feature extractor. Then, to resist the wireless noise and reconstruct the image segmentation, a semantic feature decoder and a reconstructor are designed at the receiver. Simulation results show that the proposed ISSC system can reconstruct the image segmentation accurately with a high compression ratio, and can achieve robust transmission performance against channel noise, especially at the low signal-to-noise ratio (SNR). In terms of mean Intersection over Union (mIoU), the ISSC system can achieve an increase by 75%, compared to the baselines using traditional coding method

cs.NI↗

arXiv2020

Ligand Directed Self-Assembly of Bulk Organic-Semiconductor/Quantum-Dot Blend Films Enables Near Quantitative Harvesting of Triplet Excitons

Singlet fission (SF), an exciton multiplication process occurring in organic semiconductors, offers a way to break the Shockley-Queisser limit in single-bandgap photovoltaics (PV). If the triplet excitons generated by SF can be transferred to inorganic quantum dots (QDs), where they radiatively recombine, SF based photon multiplication is achieved, converting a single high-energy photon into two low-energy photons. Such a SF photon multiplication film (SF-PMF) could raise the efficiency of the best Si-PV from 26.7% to 32.5%. But a precise nanoscale morphology is required within such a film consisting of the appropriate morphology for the organic phase, allowing for efficient SF, within which the QD emitters are well dispersed on a tens of nm length scale to enable efficient harvesting of the triplets. However, it has been a long-standing problem that the individual components in organic-QD blends have a tendency to aggregate and phase separate, due to a mismatch of their size, shape and surface energies. Here, we demonstrate a QD surface engineering approach using an electronically active, highly soluble semiconductor ligand that is matched to the SF material, which allows us to direct the self-assembly process yielding solution processed films with well-dispersed QDs and minimal aggregation, as characterised by X-ray and neutron scattering and electron microscopy. Steady state and time-resolved optical spectroscopy show that the films support efficient SF (190% yield) in the organic phase and quantitative triplet energy transfer across the organic-QD interface, resulting in 95% of the triplet excitons being harvested by the QDs. Our results establish the SF-PMF as a highly promising architecture to harness the SF process to enhance PV efficiencies, and also provide a highly versatile approach to overcome challenges in the blending of organic semiconductors with QDs.

cond-mat.mtrl-sci↗

arXiv2020

R3: A Reading Comprehension Benchmark Requiring Reasoning Processes

Existing question answering systems can only predict answers without explicit reasoning processes, which hinder their explainability and make us overestimate their ability of understanding and reasoning over natural language. In this work, we propose a novel task of reading comprehension, in which a model is required to provide final answers and reasoning processes. To this end, we introduce a formalism for reasoning over unstructured text, namely Text Reasoning Meaning Representation (TRMR). TRMR consists of three phrases, which is expressive enough to characterize the reasoning process to answer reading comprehension questions. We develop an annotation platform to facilitate TRMR's annotation, and release the R3 dataset, a \textbf{R}eading comprehension benchmark \textbf{R}equiring \textbf{R}easoning processes. R3 contains over 60K pairs of question-answer pairs and their TRMRs. Our dataset is available at: \url{http://anonymous}.

cs.CL↗