Zhijie Wang

Shuangzhi Li, Zhijie Wang, Felix Juefei-Xu et al. · mit

Object detection through LiDAR-based point cloud has recently been important in autonomous driving. Although achieving high accuracy on public benchmarks, the state-of-the-art detectors may still go wrong and cause a heavy loss due to the widespread corruptions in the real world like rain, snow, sensor noise, etc. Nevertheless, there is a lack of a large-scale dataset covering diverse scenes and realistic corruption types with different severities to develop practical and robust point cloud detectors, which is challenging due to the heavy collection costs. To alleviate the challenge and start the first step for robust point cloud detection, we propose the physical-aware simulation methods to generate degraded point clouds under different real-world common corruptions. Then, for the first attempt, we construct a benchmark based on the physical-aware common corruptions for point cloud detectors, which contains a total of 1,122,150 examples covering 7,481 scenes, 25 common corruption types, and 6 severities. With such a novel benchmark, we conduct extensive empirical studies on 8 state-of-the-art detectors that contain 6 different detection frameworks. Thus we get several insight observations revealing the vulnerabilities of the detectors and indicating the enhancement directions. Moreover, we further study the effectiveness of existing robustness enhancement methods based on data augmentation and data denoising. The benchmark can potentially be a new platform for evaluating point cloud detectors, opening a door for developing novel robustness enhancement methods.

Zhijie Wang, Yuheng Huang, Da Song et al.

Recurrent Neural Networks (RNNs) have been widely used in Natural Language Processing (NLP) tasks given its superior performance on processing sequential data. However, it is challenging to interpret and debug RNNs due to the inherent complexity and the lack of transparency of RNNs. While many explainable AI (XAI) techniques have been proposed for RNNs, most of them only support local explanations rather than global explanations. In this paper, we present DeepSeer, an interactive system that provides both global and local explanations of RNN behavior in multiple tightly-coordinated views for model understanding and debugging. The core of DeepSeer is a state abstraction method that bundles semantically similar hidden states in an RNN model and abstracts the model as a finite state machine. Users can explore the global model behavior by inspecting text patterns associated with each state and the transitions between states. Users can also dive into individual predictions by inspecting the state trace and intermediate prediction results of a given input. A between-subjects user study with 28 participants shows that, compared with a popular XAI technique, LIME, participants using DeepSeer made a deeper and more comprehensive assessment of RNN model behavior, identified the root causes of incorrect predictions more accurately, and came up with more actionable plans to improve the model performance.

Xinyu Gao, Zhijie Wang, Yang Feng et al.

Multi-Sensor Fusion (MSF) based perception systems have been the foundation in supporting many industrial applications and domains, such as self-driving cars, robotic arms, and unmanned aerial vehicles. Over the past few years, the fast progress in data-driven artificial intelligence (AI) has brought a fast-increasing trend to empower MSF systems by deep learning techniques to further improve performance, especially on intelligent systems and their perception systems. Although quite a few AI-enabled MSF perception systems and techniques have been proposed, up to the present, limited benchmarks that focus on MSF perception are publicly available. Given that many intelligent systems such as self-driving cars are operated in safety-critical contexts where perception systems play an important role, there comes an urgent need for a more in-depth understanding of the performance and reliability of these MSF systems. To bridge this gap, we initiate an early step in this direction and construct a public benchmark of AI-enabled MSF-based perception systems including three commonly adopted tasks (i.e., object detection, object tracking, and depth completion). Based on this, to comprehensively understand MSF systems' robustness and reliability, we design 14 common and realistic corruption patterns to synthesize large-scale corrupted datasets. We further perform a systematic evaluation of these systems through our large-scale evaluation. Our results reveal the vulnerability of the current AI-enabled MSF perception systems, calling for researchers and practitioners to take robustness and reliability into account when designing AI-enabled MSF.

Zhijie Wang, Yuheng Huang, Lei Ma et al.

Deep learning (DL) has become a driving force and has been widely adopted in many domains and applications with competitive performance. In practice, to solve the nontrivial and complicated tasks in real-world applications, DL is often not used standalone, but instead contributes as a piece of gadget of a larger complex AI system. Although there comes a fast increasing trend to study the quality issues of deep neural networks (DNNs) at the model level, few studies have been performed to investigate the quality of DNNs at both the unit level and the potential impacts on the system level. More importantly, it also lacks systematic investigation on how to perform the risk assessment for AI systems from unit level to system level. To bridge this gap, this paper initiates an early exploratory study of AI system risk assessment from both the data distribution and uncertainty angles to address these issues. We propose a general framework with an exploratory study for analyzing AI systems. After large-scale (700+ experimental configurations and 5000+ GPU hours) experiments and in-depth investigations, we reached a few key interesting findings that highlight the practical need and opportunities for more in-depth investigations into AI systems.

Yuheng Huang, Jiayang Song, Zhijie Wang et al.

The recent performance leap of Large Language Models (LLMs) opens up new opportunities across numerous industrial applications and domains. However, erroneous generations, such as false predictions, misinformation, and hallucination made by LLMs, have also raised severe concerns for the trustworthiness of LLMs', especially in safety-, security- and reliability-sensitive scenarios, potentially hindering real-world adoptions. While uncertainty estimation has shown its potential for interpreting the prediction risks made by general machine learning (ML) models, little is known about whether and to what extent it can help explore an LLM's capabilities and counteract its undesired behavior. To bridge the gap, in this paper, we initiate an exploratory study on the risk assessment of LLMs from the lens of uncertainty. In particular, we experiment with twelve uncertainty estimation methods and four LLMs on four prominent natural language processing (NLP) tasks to investigate to what extent uncertainty estimation techniques could help characterize the prediction risks of LLMs. Our findings validate the effectiveness of uncertainty estimation for revealing LLMs' uncertain/non-factual predictions. In addition to general NLP tasks, we extensively conduct experiments with four LLMs for code generation on two datasets. We find that uncertainty estimation can potentially uncover buggy programs generated by LLMs. Insights from our study shed light on future design and development for reliable LLMs, facilitating further research toward enhancing the trustworthiness of LLMs.

Zhijie Wang, Masanori Suganuma, Takayuki Okatani

Unsupervised domain adaptation (UDA) adapts a model trained on one domain (called source) to a novel domain (called target) using only unlabeled data. Due to its high annotation cost, researchers have developed many UDA methods for semantic segmentation, which assume no labeled sample is available in the target domain. We question the practicality of this assumption for two reasons. First, after training a model with a UDA method, we must somehow verify the model before deployment. Second, UDA methods have at least a few hyper-parameters that need to be determined. The surest solution to these is to evaluate the model using validation data, i.e., a certain amount of labeled target-domain samples. This question about the basic assumption of UDA leads us to rethink UDA from a data-centric point of view. Specifically, we assume we have access to a minimum level of labeled data. Then, we ask how much is necessary to find good hyper-parameters of existing UDA methods. We then consider what if we use the same data for supervised training of the same model, e.g., finetuning. We conducted experiments to answer these questions with popular scenarios, {GTA5, SYNTHIA}$\rightarrow$Cityscapes. We found that i) choosing good hyper-parameters needs only a few labeled images for some UDA methods whereas a lot more for others; and ii) simple finetuning works surprisingly well; it outperforms many UDA methods if only several dozens of labeled images are available.

Bonan Kou, Shengmai Chen, Zhijie Wang et al.

Large Language Models (LLMs) have recently been widely used for code generation. Due to the complexity and opacity of LLMs, little is known about how these models generate code. We made the first attempt to bridge this knowledge gap by investigating whether LLMs attend to the same parts of a task description as human programmers during code generation. An analysis of six LLMs, including GPT-4, on two popular code generation benchmarks revealed a consistent misalignment between LLMs' and programmers' attention. We manually analyzed 211 incorrect code snippets and found five attention patterns that can be used to explain many code generation errors. Finally, a user study showed that model attention computed by a perturbation-based method is often favored by human programmers. Our findings highlight the need for human-aligned LLMs for better interpretability and programmer trust.

Da Song, Zhijie Wang, Yuheng Huang et al.

Machine Learning (ML) has been widely used in Natural Language Processing (NLP) applications. A fundamental assumption in ML is that training data and real-world data should follow a similar distribution. However, a deployed ML model may suffer from out-of-distribution (OOD) issues due to distribution shifts in the real-world data. Though many algorithms have been proposed to detect OOD data from text corpora, there is still a lack of interactive tool support for ML developers. In this work, we propose DeepLens, an interactive system that helps users detect and explore OOD issues in massive text corpora. Users can efficiently explore different OOD types in DeepLens with the help of a text clustering method. Users can also dig into a specific text by inspecting salient words highlighted through neuron activation analysis. In a within-subjects user study with 24 participants, participants using DeepLens were able to find nearly twice more types of OOD issues accurately with 22% more confidence compared with a variant of DeepLens that has no interaction or visualization support.

Zhijie Wang, Yuheng Huang, Da Song et al.

The recent advancements in Generative AI have significantly advanced the field of text-to-image generation. The state-of-the-art text-to-image model, Stable Diffusion, is now capable of synthesizing high-quality images with a strong sense of aesthetics. Crafting text prompts that align with the model's interpretation and the user's intent thus becomes crucial. However, prompting remains challenging for novice users due to the complexity of the stable diffusion model and the non-trivial efforts required for iteratively editing and refining the text prompts. To address these challenges, we propose PromptCharm, a mixed-initiative system that facilitates text-to-image creation through multi-modal prompt engineering and refinement. To assist novice users in prompting, PromptCharm first automatically refines and optimizes the user's initial prompt. Furthermore, PromptCharm supports the user in exploring and selecting different image styles within a large database. To assist users in effectively refining their prompts and images, PromptCharm renders model explanations by visualizing the model's attention values. If the user notices any unsatisfactory areas in the generated images, they can further refine the images through model attention adjustment or image inpainting within the rich feedback loop of PromptCharm. To evaluate the effectiveness and usability of PromptCharm, we conducted a controlled user study with 12 participants and an exploratory user study with another 12 participants. These two studies show that participants using PromptCharm were able to create images with higher quality and better aligned with the user's expectations compared with using two variants of PromptCharm that lacked interaction or visualization support.

Shengming Zhao, Yuchen Shao, Yuheng Huang et al.

Retrieval-Augmented Generation (RAG) has emerged as a critical technique for enhancing large language model (LLM) capabilities. However, practitioners face significant challenges when making RAG deployment decisions. While existing research prioritizes algorithmic innovations, a systematic gap persists in understanding fundamental engineering trade-offs that determine RAG success. We present the first comprehensive study of three universal RAG deployment decisions: whether to deploy RAG, how much information to retrieve, and how to integrate retrieved knowledge effectively. Through systematic experiments across three LLMs and six datasets spanning question answering and code generation tasks, we reveal critical insights: (1) RAG deployment must be highly selective, with variable recall thresholds and failure modes affecting up to 12.6\% of samples even with perfect documents. (2) Optimal retrieval volume exhibits task-dependent behavior QA tasks show universal patterns (5-10 documents optimal) while code generation requires scenario-specific optimization. (3) Knowledge integration effectiveness depends on task and model characteristics, with code generation benefiting significantly from prompting methods while question answering shows minimal improvement. These findings demonstrate that universal RAG strategies prove inadequate. Effective RAG systems require context-aware design decisions based on task characteristics and model capabilities. Our analysis provides evidence-based guidance for practitioners and establishes foundational insights for principled RAG deployment.

Hua Qi, Zhijie Wang, Qing Guo et al.

Over the past few years, deep neural networks (DNNs) have achieved tremendous success and have been continuously applied in many application domains. However, during the practical deployment in the industrial tasks, DNNs are found to be erroneous-prone due to various reasons such as overfitting, lacking robustness to real-world corruptions during practical usage. To address these challenges, many recent attempts have been made to repair DNNs for version updates under practical operational contexts by updating weights (i.e., network parameters) through retraining, fine-tuning, or direct weight fixing at a neural level. In this work, as the first attempt, we initiate to repair DNNs by jointly optimizing the architecture and weights at a higher (i.e., block) level. We first perform empirical studies to investigate the limitation of whole network-level and layer-level repairing, which motivates us to explore a novel repairing direction for DNN repair at the block level. To this end, we first propose adversarial-aware spectrum analysis for vulnerable block localization that considers the neurons' status and weights' gradients in blocks during the forward and backward processes, which enables more accurate candidate block localization for repairing even under a few examples. Then, we further propose the architecture-oriented search-based repairing that relaxes the targeted block to a continuous repairing search space at higher deep feature levels. By jointly optimizing the architecture and weights in that space, we can identify a much better block architecture. We implement our proposed repairing techniques as a tool, named ArchRepair, and conduct extensive experiments to validate the proposed method. The results show that our method can not only repair but also enhance accuracy & robustness, outperforming the state-of-the-art DNN repair techniques.

Jiayang Song, Deyun Lyu, Zhenya Zhang et al.

Cyber-physical systems (CPS) have been broadly deployed in safety-critical domains, such as automotive systems, avionics, medical devices, etc. In recent years, Artificial Intelligence (AI) has been increasingly adopted to control CPS. Despite the popularity of AI-enabled CPS, few benchmarks are publicly available. There is also a lack of deep understanding on the performance and reliability of AI-enabled CPS across different industrial domains. To bridge this gap, we initiate to create a public benchmark of industry-level CPS in seven domains and build AI controllers for them via state-of-the-art deep reinforcement learning (DRL) methods. Based on that, we further perform a systematic evaluation of these AI-enabled systems with their traditional counterparts to identify the current challenges and explore future opportunities. Our key findings include (1) AI controllers do not always outperform traditional controllers, (2) existing CPS testing techniques (falsification, specifically) fall short of analyzing AI-enabled CPS, and (3) building a hybrid system that strategically combines and switches between AI controllers and traditional controllers can achieve better performance across different domains. Our results highlight the need for new testing techniques for AI-enabled CPS and the need for more investigations into hybrid CPS systems to achieve optimal performance and reliability.

Zhijie Wang, Masanori Suganuma, Takayuki Okatani

This study is concerned with few-shot segmentation, i.e., segmenting the region of an unseen object class in a query image, given support image(s) of its instances. The current methods rely on the pretrained CNN features of the support and query images. The key to good performance depends on the proper fusion of their mid-level and high-level features; the former contains shape-oriented information, while the latter has class-oriented information. Current state-of-the-art methods follow the approach of Tian et al., which gives the mid-level features the primary role and the high-level features the secondary role. In this paper, we reinterpret this widely employed approach by redifining the roles of the multi-level features; we swap the primary and secondary roles. Specifically, we regard that the current methods improve the initial estimate generated from the high-level features using the mid-level features. This reinterpretation suggests a new application of the current methods: to apply the same network multiple times to iteratively update the estimate of the object's region, starting from its initial estimate. Our experiments show that this method is effective and has updated the previous state-of-the-art on COCO-20$^i$ in the 1-shot and 5-shot settings and on PASCAL-5$^i$ in the 1-shot setting.

Zhijie Wang, Xing Liu, Masanori Suganuma et al.

Semantic segmentation requires a lot of training data, which necessitates costly annotation. There have been many studies on unsupervised domain adaptation (UDA) from one domain to another, e.g., from computer graphics to real images. However, there is still a gap in accuracy between UDA and supervised training on native domain data. It is arguably attributable to class-level misalignment between the source and target domain data. To cope with this, we propose a method that applies adversarial training to align two feature distributions in the target domain. It uses a self-training framework to split the image into two regions (i.e., trusted and untrusted), which form two distributions to align in the feature space. We term this approach cross-region adaptation (CRA) to distinguish from the previous methods of aligning different domain distributions, which we call cross-domain adaptation (CDA). CRA can be applied after any CDA method. Experimental results show that this always improves the accuracy of the combined CDA method, having updated the state-of-the-art.

Xiao Wang, Zhijie Wang, Yolande M. Pengetnze et al.

Pediatric asthma is the most prevalent chronic childhood illness, afflicting about 6.2 million children in the United States. However, asthma could be better managed by identifying and avoiding triggers, educating about medications and proper disease management strategies. This research utilizes deep learning methodologies to predict asthma-related emergency department (ED) visit within 3 months using Medicaid claims data. We compare prediction results against traditional statistical classification model - penalized Lasso logistic regression, which we trained and have deployed since 2015. The results have indicated that deep learning model Artificial Neural Networks (ANN) slightly outperforms (with AUC = 0.845) the Lasso logistic regression (with AUC = 0.842). The reason may come from the nonlinear nature of ANN.