arXiv subjects

Chen Tang

Publications and source records attributed to Chen Tang.

140 records · Page 8Linked to original sources

arXiv2022

Hierarchical Planning Through Goal-Conditioned Offline Reinforcement Learning

Offline Reinforcement learning (RL) has shown potent in many safe-critical tasks in robotics where exploration is risky and expensive. However, it still struggles to acquire skills in temporally extended tasks. In this paper, we study the problem of offline RL for temporally extended tasks. We propose a hierarchical planning framework, consisting of a low-level goal-conditioned RL policy and a high-level goal planner. The low-level policy is trained via offline RL. We improve the offline training to deal with out-of-distribution goals by a perturbed goal sampling process. The high-level planner selects intermediate sub-goals by taking advantages of model-based planning methods. It plans over future sub-goal sequences based on the learned value function of the low-level policy. We adopt a Conditional Variational Autoencoder to sample meaningful high-dimensional sub-goal candidates and to solve the high-level long-term strategy optimization problem. We evaluate our proposed method in long-horizon driving and robot navigation tasks. Experiments show that our method outperforms baselines with different hierarchical designs and other regular planners without hierarchy in these complex tasks.

cs.RO↗

arXiv2022

Arbitrary Bit-width Network: A Joint Layer-Wise Quantization and Adaptive Inference Approach

Conventional model quantization methods use a fixed quantization scheme to different data samples, which ignores the inherent "recognition difficulty" differences between various samples. We propose to feed different data samples with varying quantization schemes to achieve a data-dependent dynamic inference, at a fine-grained layer level. However, enabling this adaptive inference with changeable layer-wise quantization schemes is challenging because the combination of bit-widths and layers is growing exponentially, making it extremely difficult to train a single model in such a vast searching space and use it in practice. To solve this problem, we present the Arbitrary Bit-width Network (ABN), where the bit-widths of a single deep network can change at runtime for different data samples, with a layer-wise granularity. Specifically, first we build a weight-shared layer-wise quantizable "super-network" in which each layer can be allocated with multiple bit-widths and thus quantized differently on demand. The super-network provides a considerably large number of combinations of bit-widths and layers, each of which can be used during inference without retraining or storing myriad models. Second, based on the well-trained super-network, each layer's runtime bit-width selection decision is modeled as a Markov Decision Process (MDP) and solved by an adaptive inference strategy accordingly. Experiments show that the super-network can be built without accuracy degradation, and the bit-widths allocation of each layer can be adjusted to deal with various inputs on the fly. On ImageNet classification, we achieve 1.1% top1 accuracy improvement while saving 36.2% BitOps.

cs.CV↗

arXiv2022

PreTraM: Self-Supervised Pre-training via Connecting Trajectory and Map

Deep learning has recently achieved significant progress in trajectory forecasting. However, the scarcity of trajectory data inhibits the data-hungry deep-learning models from learning good representations. While mature representation learning methods exist in computer vision and natural language processing, these pre-training methods require large-scale data. It is hard to replicate these approaches in trajectory forecasting due to the lack of adequate trajectory data (e.g., 34K samples in the nuScenes dataset). To work around the scarcity of trajectory data, we resort to another data modality closely related to trajectories-HD-maps, which is abundantly provided in existing datasets. In this paper, we propose PreTraM, a self-supervised pre-training scheme via connecting trajectories and maps for trajectory forecasting. Specifically, PreTraM consists of two parts: 1) Trajectory-Map Contrastive Learning, where we project trajectories and maps to a shared embedding space with cross-modal contrastive learning, and 2) Map Contrastive Learning, where we enhance map representation with contrastive learning on large quantities of HD-maps. On top of popular baselines such as AgentFormer and Trajectron++, PreTraM boosts their performance by 5.5% and 6.9% relatively in FDE-10 on the challenging nuScenes dataset. We show that PreTraM improves data efficiency and scales well with model size.

cs.CV↗

arXiv2022

Clustering and Forecasting Multiple Functional Time Series

Modelling and forecasting homogeneous age-specific mortality rates of multiple countries could lead to improvements in long-term forecasting. Data fed into joint models are often grouped according to nominal attributes, such as geographic regions, ethnic groups, and socioeconomic status, which may still contain heterogeneity and deteriorate the forecast results. Our paper proposes a novel clustering technique to pursue homogeneity among multiple functional time series based on functional panel data modelling to address this issue. Using a functional panel data model with fixed effects, we can extract common functional time series features. These common features could be decomposed into two components: the functional time trend and the mode of variations of functions (functional pattern). The functional time trend reflects the dynamics across time, while the functional pattern captures the fluctuations within curves. The proposed clustering method searches for homogeneous age-specific mortality rates of multiple countries by accounting for both the modes of variations and the temporal dynamics among curves. We demonstrate that the proposed clustering technique outperforms other existing methods through a Monte Carlo simulation and could handle complicated cases with slow decaying eigenvalues. In empirical data analysis, we find that the clustering results of age-specific mortality rates can be explained by the combination of geographic region, ethnic groups, and socioeconomic status. We further show that our model produces more accurate forecasts than several benchmark methods in forecasting age-specific mortality rates.

stat.ME↗

arXiv2021

Exploring Social Posterior Collapse in Variational Autoencoder for Interaction Modeling

Multi-agent behavior modeling and trajectory forecasting are crucial for the safe navigation of autonomous agents in interactive scenarios. Variational Autoencoder (VAE) has been widely applied in multi-agent interaction modeling to generate diverse behavior and learn a low-dimensional representation for interacting systems. However, existing literature did not formally discuss if a VAE-based model can properly encode interaction into its latent space. In this work, we argue that one of the typical formulations of VAEs in multi-agent modeling suffers from an issue we refer to as social posterior collapse, i.e., the model is prone to ignoring historical social context when predicting the future trajectory of an agent. It could cause significant prediction errors and poor generalization performance. We analyze the reason behind this under-explored phenomenon and propose several measures to tackle it. Afterward, we implement the proposed framework and experiment on real-world datasets for multi-agent trajectory prediction. In particular, we propose a novel sparse graph attention message-passing (sparse-GAMP) layer, which helps us detect social posterior collapse in our experiments. In the experiments, we verify that social posterior collapse indeed occurs. Also, the proposed measures are effective in alleviating the issue. As a result, the model attains better generalization performance when historical social context is informative for prediction.

cs.LG↗

arXiv2021

Dealing with the Unknown: Pessimistic Offline Reinforcement Learning

Reinforcement Learning (RL) has been shown effective in domains where the agent can learn policies by actively interacting with its operating environment. However, if we change the RL scheme to offline setting where the agent can only update its policy via static datasets, one of the major issues in offline reinforcement learning emerges, i.e. distributional shift. We propose a Pessimistic Offline Reinforcement Learning (PessORL) algorithm to actively lead the agent back to the area where it is familiar by manipulating the value function. We focus on problems caused by out-of-distribution (OOD) states, and deliberately penalize high values at states that are absent in the training dataset, so that the learned pessimistic value function lower bounds the true value anywhere within the state space. We evaluate the PessORL algorithm on various benchmark tasks, where we show that our method gains better performance by explicitly handling OOD states, when compared to those methods merely considering OOD actions.

cs.LG↗

arXiv2021

Equilibration of the planar modes of ultracold two dimensional ion crystals in a Penning trap

Planar thermal equilibration is studied using direct numerical simulations of ultracold two-dimensional (2D) ion crystals in a Penning trap with a rotating wall. The large magnetic field of the trap splits the modes that describe in-plane motion of the ions into two branches: High frequency cyclotron modes dominated by kinetic energy and low frequency $\mathbf{E \times B}$ modes dominated by potential energy associated with thermal position displacements. Using an eigenmode analysis we extract the equilibration rate between these two branches as a function of the ratio of the frequencies that characterize the two branches and observe this equilibration rate to be exponentially suppressed as the ratio increases. Under experimental conditions relevant for current work at NIST, the predicted equilibration time is orders of magnitude longer than any relevant experimental timescales. We also study the coupling rate dependence on the thermal temperature and the number of ions. Besides, we show how increasing the rotating wall strength improves crystal stability. These details of in-plane mode dynamics help set the stage for developing strategies to efficiently cool the in-plane modes and improve the performance of single-plane ion crystals for quantum information processing.

physics.plasm-ph↗

arXiv2021

Adaptive Pixel-wise Structured Sparse Network for Efficient CNNs

To accelerate deep CNN models, this paper proposes a novel spatially adaptive framework that can dynamically generate pixel-wise sparsity according to the input image. The sparse scheme is pixel-wise refined, regional adaptive under a unified importance map, which makes it friendly to hardware implementation. A sparse controlling method is further presented to enable online adjustment for applications with different precision/latency requirements. The sparse model is applicable to a wide range of vision tasks. Experimental results show that this method efficiently improve the computing efficiency for both image classification using ResNet-18 and super resolution using SRResNet. On image classification task, our method can save 30%-70% MACs with a slightly drop in top-1 and top-5 accuracy. On super resolution task, our method can reduce more than 90% MACs while only causing around 0.1 dB and 0.01 decreasing in PSNR and SSIM. Hardware validation is also included.

cs.CV↗

arXiv2020

Functional data analysis: An application to COVID-19 data in the United States

The COVID-19 pandemic so far has caused huge negative impacts on different areas all over the world, and the United States (US) is one of the most affected countries. In this paper, we use methods from the functional data analysis to look into the COVID-19 data in the US. We explore the modes of variation of the data through a functional principal component analysis (FPCA), and study the canonical correlation between confirmed and death cases. In addition, we run a cluster analysis at the state level so as to investigate the relation between geographical locations and the clustering structure. Lastly, we consider a functional time series model fitted to the cumulative confirmed cases in the US, and make forecasts based on the dynamic FPCA. Both point and interval forecasts are provided, and the methods for assessing the accuracy of the forecasts are also included.

stat.AP↗

arXiv2020

Broadening of the drumhead mode spectrum due to in-plane thermal fluctuations of two-dimensional trapped ion crystals in a Penning trap

Two-dimensional crystals of ions stored in Penning traps are a leading platform for quantum simulation and sensing experiments. For small amplitudes, the out-of-plane motion of such crystals can be described by a discrete set of normal modes called the drumhead modes, which can be used to implement a range of quantum information protocols. However, experimental observations of crystals with Doppler-cooled and even near-ground-state-cooled drumhead modes reveal an unresolved drumhead mode spectrum. In this work, we establish in-plane thermal fluctuations in ion positions as a major contributor to the broadening of the drumhead mode spectrum. In the process, we demonstrate how the confining magnetic field leads to unconventional in-plane normal modes, whose average potential and kinetic energies are not equal. This property, in turn, has implications for the sampling procedure required to choose the in-plane initial conditions for molecular dynamics simulations. For current operating conditions of the NIST Penning trap, our study suggests that the two dimensional crystals produced in this trap undergo in-plane potential energy fluctuations of the order of $10$ mK. Our study therefore motivates the need for designing improved techniques to cool the in-plane degrees of freedom.

quant-ph↗

arXiv2019

Adaptive Probabilistic Vehicle Trajectory Prediction Through Physically Feasible Bayesian Recurrent Neural Network

Probabilistic vehicle trajectory prediction is essential for robust safety of autonomous driving. Current methods for long-term trajectory prediction cannot guarantee the physical feasibility of predicted distribution. Moreover, their models cannot adapt to the driving policy of the predicted target human driver. In this work, we propose to overcome these two shortcomings by a Bayesian recurrent neural network model consisting of Bayesian-neural-network-based policy model and known physical model of the scenario. Bayesian neural network can ensemble complicated output distribution, enabling rich family of trajectory distribution. The embedded physical model ensures feasibility of the distribution. Moreover, the adopted gradient-based training method allows direct optimization for better performance in long prediction horizon. Furthermore, a particle-filter-based parameter adaptation algorithm is designed to adapt the policy Bayesian neural network to the predicted target online. Effectiveness of the proposed methods is verified with a toy example with multi-modal stochastic feedback gain and naturalistic car following data.

cs.LG↗

arXiv2019

ZPD Teaching Strategies for Deep Reinforcement Learning from Demonstrations

Learning from demonstrations is a popular tool for accelerating and reducing the exploration requirements of reinforcement learning. When providing expert demonstrations to human students, we know that the demonstrations must fall within a particular range of difficulties called the "Zone of Proximal Development (ZPD)". If they are too easy the student learns nothing, but if they are too difficult the student is unable to follow along. This raises the question: Given a set of potential demonstrators, which among them is best suited for teaching any particular learner? Prior work, such as the popular Deep Q-learning from Demonstrations (DQfD) algorithm has generally focused on single demonstrators. In this work we consider the problem of choosing among multiple demonstrators of varying skill levels. Our results align with intuition from human learners: it is not always the best policy to draw demonstrations from the best performing demonstrator (in terms of reward). We show that careful selection of teaching strategies can result in sample efficiency gains in the learner's environment across nine Atari games

cs.LG↗

arXiv2019

First principle simulation of ultra-cold ion crystals in a Penning trap with Doppler cooling and a rotating wall potential

A direct numerical simulation of many interacting ions in a Penning trap with a rotating wall is presented. The ion dynamics is modelled classically. Both axial and planar Doppler laser cooling are modeled using stochastic momentum impulses based on two-level atomic scattering rates. The plasmas being modeled are ultra-cold two-dimensional crystals made up of 100's of ions. We compare Doppler cooled results directly to a previous linear eigenmodes analysis. Agreement in both frequency and mode structure are obtained. Additionally, when Doppler laser cooling is applied, the laser cooled steady state plasma axial temperature agrees with the Doppler cooling limit. Numerical simulations using the approach described and benchmarked here will provide insights into the dynamics of large trapped-ion crystals, improving their performance as a platform for quantum simulation and sensing.

physics.plasm-ph↗

arXiv2018

Zero-shot Deep Reinforcement Learning Driving Policy Transfer for Autonomous Vehicles based on Robust Control

Although deep reinforcement learning (deep RL) methods have lots of strengths that are favorable if applied to autonomous driving, real deep RL applications in autonomous driving have been slowed down by the modeling gap between the source (training) domain and the target (deployment) domain. Unlike current policy transfer approaches, which generally limit to the usage of uninterpretable neural network representations as the transferred features, we propose to transfer concrete kinematic quantities in autonomous driving. The proposed robust-control-based (RC) generic transfer architecture, which we call RL-RC, incorporates a transferable hierarchical RL trajectory planner and a robust tracking controller based on disturbance observer (DOB). The deep RL policies trained with known nominal dynamics model are transfered directly to the target domain, DOB-based robust tracking control is applied to tackle the modeling gap including the vehicle dynamics errors and the external disturbances such as side forces. We provide simulations validating the capability of the proposed method to achieve zero-shot transfer across multiple driving scenarios such as lane keeping, lane changing and obstacle avoidance.

cs.LG↗