Wencan Zhang

Hao Ye, Jisheng Dang, Junfeng Fang et al.

Recent extensive research has demonstrated that the enhanced reasoning capabilities acquired by models through Reinforcement Learning with Verifiable Rewards (RLVR) are primarily concentrated within the rank-1 components. Predicated on this observation, we employed Periodic Rank-1 Substitution and identified a counterintuitive phenomenon: RLVR may exhibit implicit reward overfitting to the training dataset. Specifically, the model can achieve satisfactory performance on the test set even when its rewards remain relatively low during the training process. Furthermore, we characterize three distinct properties of RL training: (1) The effective rank-1 component in RLVR don't maintain other model knowledge except mathematical reasoning capability. (2) RLVR fundamentally functions by optimizing a specific singular spectrum. The distribution of singular values of almost all linear layers in RLVR-trained model behaves like heavy-tailed distribution. (3) the left singular vectors associated with rank-1 components demonstrate a stronger alignment tendency during training, which echoes the discovery that RLVR is optimizing sampling efficiency in essence. Taken together, our findings and analysis further reveal how RLVR shapes model parameters and offer potential insights for improving existing RL paradigms or other training paradigms to implement continual learning.

Haoyang Chen, Peiyan Sun, Qiyuan Song et al.

Ride-hailing platforms have been facing the challenge of balancing demand and supply. Existing vehicle reposition techniques often treat drivers as homogeneous agents and relocate them deterministically, assuming compliance with the reposition. In this paper, we consider a more realistic and driver-centric scenario where drivers have unique cruising preferences and can decide whether to take the recommendation or not on their own. We propose i-Rebalance, a personalized vehicle reposition technique with deep reinforcement learning (DRL). i-Rebalance estimates drivers' decisions on accepting reposition recommendations through an on-field user study involving 99 real drivers. To optimize supply-demand balance and enhance preference satisfaction simultaneously, i-Rebalance has a sequential reposition strategy with dual DRL agents: Grid Agent to determine the reposition order of idle vehicles, and Vehicle Agent to provide personalized recommendations to each vehicle in the pre-defined order. This sequential learning strategy facilitates more effective policy training within a smaller action space compared to traditional joint-action methods. Evaluation of real-world trajectory data shows that i-Rebalance improves driver acceptance rate by 38.07% and total driver income by 9.97%.

Wencan Zhang, Mariella Dimiccoli, Brian Y. Lim

Model explanations such as saliency maps can improve user trust in AI by highlighting important features for a prediction. However, these become distorted and misleading when explaining predictions of images that are subject to systematic error (bias). Furthermore, the distortions persist despite model fine-tuning on images biased by different factors (blur, color temperature, day/night). We present Debiased-CAM to recover explanation faithfulness across various bias types and levels by training a multi-input, multi-task model with auxiliary tasks for explanation and bias level predictions. In simulation studies, the approach not only enhanced prediction accuracy, but also generated highly faithful explanations about these predictions as if the images were unbiased. In user studies, debiased explanations improved user task performance, perceived truthfulness and perceived helpfulness. Debiased training can provide a versatile platform for robust performance and explanation faithfulness for a wide range of applications with data biases.

Wencan Zhang, Brian Y. Lim

Machine learning models need to provide contrastive explanations, since people often seek to understand why a puzzling prediction occurred instead of some expected outcome. Current contrastive explanations are rudimentary comparisons between examples or raw features, which remain difficult to interpret, since they lack semantic meaning. We argue that explanations must be more relatable to other concepts, hypotheticals, and associations. Inspired by the perceptual process from cognitive psychology, we propose the XAI Perceptual Processing Framework and RexNet model for relatable explainable AI with Contrastive Saliency, Counterfactual Synthetic, and Contrastive Cues explanations. We investigated the application of vocal emotion recognition, and implemented a modular multi-task deep neural network to predict and explain emotions from speech. From think-aloud and controlled studies, we found that counterfactual explanations were useful and further enhanced with semantic cues, but not saliency explanations. This work provides insights into providing and evaluating relatable contrastive explainable AI for perception applications.

Xuejun Zhao, Wencan Zhang, Xiaokui Xiao et al.

The successful deployment of artificial intelligence (AI) in many domains from healthcare to hiring requires their responsible use, particularly in model explanations and privacy. Explainable artificial intelligence (XAI) provides more information to help users to understand model decisions, yet this additional knowledge exposes additional risks for privacy attacks. Hence, providing explanation harms privacy. We study this risk for image-based model inversion attacks and identified several attack architectures with increasing performance to reconstruct private image data from model explanations. We have developed several multi-modal transposed CNN architectures that achieve significantly higher inversion performance than using the target model prediction only. These XAI-aware inversion models were designed to exploit the spatial knowledge in image explanations. To understand which explanations have higher privacy risk, we analyzed how various explanation types and factors influence inversion performance. In spite of some models not providing explanations, we further demonstrate increased inversion performance even for non-explainable target models by exploiting explanations of surrogate models through attention transfer. This method first inverts an explanation from the target prediction, then reconstructs the target image. These threats highlight the urgent and significant privacy risks of explanations and calls attention for new privacy preservation techniques that balance the dual-requirement for AI explainability and privacy.

Danding Wang, Wencan Zhang, Brian Y. Lim

Feature attribution is widely used in interpretable machine learning to explain how influential each measured input feature value is for an output inference. However, measurements can be uncertain, and it is unclear how the awareness of input uncertainty can affect the trust in explanations. We propose and study two approaches to help users to manage their perception of uncertainty in a model explanation: 1) transparently show uncertainty in feature attributions to allow users to reflect on, and 2) suppress attribution to features with uncertain measurements and shift attribution to other features by regularizing with an uncertainty penalty. Through simulation experiments, qualitative interviews, and quantitative user evaluations, we identified the benefits of moderately suppressing attribution uncertainty, and concerns regarding showing attribution uncertainty. This work adds to the understanding of handling and communicating uncertainty for model interpretability.

Wencan Zhang, Mariella Dimiccoli, Brian Y. Lim

Model explanations such as saliency maps can improve user trust in AI by highlighting important features for a prediction. However, these become distorted and misleading when explaining predictions of images that are subject to systematic error (bias) by perturbations and corruptions. Furthermore, the distortions persist despite model fine-tuning on images biased by different factors (blur, color temperature, day/night). We present Debiased-CAM to recover explanation faithfulness across various bias types and levels by training a multi-input, multi-task model with auxiliary tasks for explanation and bias level predictions. In simulation studies, the approach not only enhanced prediction accuracy, but also generated highly faithful explanations about these predictions as if the images were unbiased. In user studies, debiased explanations improved user task performance, perceived truthfulness and perceived helpfulness. Debiased training can provide a versatile platform for robust performance and explanation faithfulness for a wide range of applications with data biases.