Uncertainty quantification (UQ) enhances the reliability of Large Language Models (LLMs) by estimating confidence in outputs, enabling risk mitigation and selective prediction. However, traditional UQ methods struggle with LLMs due to computational constraints and decoding inconsistencies. Moreover, LLMs introduce unique uncertainty sources, such as input ambiguity, reasoning path divergence, and decoding stochasticity, that extend beyond classical aleatoric and epistemic uncertainty. To address this, we introduce a new taxonomy that categorizes UQ methods based on computational efficiency and uncertainty dimensions, including input, reasoning, parameter, and prediction uncertainty. We evaluate existing techniques, summarize existing benchmarks and metrics for UQ, assess their real-world applicability, and identify open challenges, emphasizing the need for scalable, interpretable, and robust UQ approaches to enhance LLM reliability.

Response generation is one of the critical components in task-oriented dialog systems.Existing studies have shown that large pre-trained language models can be adapted to this task.The typical paradigm of adapting such extremely large language models would be by fine-tuning on the downstream tasks which is not only time-consuming but also involves significant resources and access to fine-tuning data.Prompting (Schick and Schütze, 2020) has been an alternative to fine-tuning in many NLP tasks.In our work, we explore the idea of using prompting for response generation in task-oriented dialog systems.Specifically, we propose an approach that performs contextual dynamic prompting where the prompts are learnt from dialog contexts.We aim to distill useful prompting signals from the dialog context.On experiments with MultiWOZ 2.2 dataset (Zang et al., 2020), we show that contextual dynamic prompts improve response generation in terms of combined score (Mehri et al., 2019a) by 3 absolute points, and a massive 20 points when dialog states are incorporated.Furthermore, human annotation on these conversations found that agents which incorporate context were preferred over agents with vanilla prefix-tuning.

While a vast collection of explainable AI (XAI) algorithms has been developed in recent years, they have been criticized for significant gaps with how humans produce and consume explanations. As a result, current XAI techniques are often found to be hard to use and lack effectiveness. In this work, we attempt to close these gaps by making AI explanations selective ---a fundamental property of human explanations---by selectively presenting a subset of model reasoning based on what aligns with the recipient's preferences. We propose a general framework for generating selective explanations by leveraging human input on a small dataset. This framework opens up a rich design space that accounts for different selectivity goals, types of input, and more. As a showcase, we use a decision-support task to explore selective explanations based on what the decision-maker would consider relevant to the decision task. We conducted two experimental studies to examine three paradigms based on our proposed framework: in Study 1, we ask the participants to provide critique-based or open-ended input to generate selective explanations (self-input). In Study 2, we show the participants selective explanations based on input from a panel of similar users (annotator input). Our experiments demonstrate the promise of selective explanations in reducing over-reliance on AI and improving collaborative decision making and subjective perceptions of the AI system, but also paint a nuanced picture that attributes some of these positive effects to the opportunity to provide one's own input to augment AI explanations. Overall, our work proposes a novel XAI framework inspired by human communication behaviors and demonstrates its potential to encourage future work to make AI explanations more human-compatible.

Chi-Yang Hsu, Yun-Wei Chu, Vincent Chen, Kuan-Chieh Lo, Chacha Chen, Ting-Hao Huang, Lun-Wei Ku. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Understanding city-wide traffic problems may benefit many downstream applications, such as city planning and public transportation development. One key step to understand traffic is to reveal how many people travel from one location to another during one period (we call TOD, short for temporal origin-destination). With TOD, we can rebuild the city-wide traffic by simulating the volume and speed on each road segment.Frequently used mobility data, e.g., GPS trajectories, surveillance cameras, can only cover a subset of vehicles or selected regions of the city. Hence, we propose to use pervasive speed data to recover TOD, and use other mobility data as auxiliary data. To the best of our knowledge, we are the first to work on this challenging problem. It is highly challenging because the speed is generated from a complex process from TOD, and there exists multiple TOD distributions that may generate similar city-wide road speed observations. We propose a new method that models the complex process via separate modules and takes auxiliary data to eliminate infeasible solutions. Extensive experiments on synthetic and real datasets have shown the superior performance of our model over baselines.

Small data has been a barrier for many machine learning tasks, especially when applied in scientific domains. Fortunately, we can utilize domain knowledge to make up the lack of data. Hence, in this paper, we propose a hybrid model KRL that treats domain knowledge model as a weak learner and uses another neural net model to boost it. We prove that KRL is guaranteed to improve over pure domain knowledge model and pure neural net model under certain loss functions. Extensive experiments have shown the superior performance of KRL over baselines. In addition, several case studies have explained how the domain knowledge can assist the prediction.

Successful health risk prediction demands accuracy and reliability of the model. Existing predictive models mainly depend on mining electronic health records (EHR) with advanced deep learning techniques to improve model accuracy. However, they all ignore the importance of publicly available online health data, especially socioeconomic status, environmental factors, and detailed demographic information for each location, which are all strong predictive signals and can definitely augment precision medicine. To achieve model reliability, the model needs to provide accurate prediction and uncertainty score of the prediction. However, existing uncertainty estimation approaches often failed in handling high-dimensional data, which are present in multi-sourced data.

Traffic congestion plagues cities around the world. Recent years have witnessed an unprecedented trend in applying reinforcement learning for traffic signal control. However, the primary challenge is to control and coordinate traffic lights in large-scale urban networks. No one has ever tested RL models on a network of more than a thousand traffic lights. In this paper, we tackle the problem of multi-intersection traffic signal control, especially for large-scale networks, based on RL techniques and transportation theories. This problem is quite difficult because there are challenges such as scalability, signal coordination, data feasibility, etc. To address these challenges, we (1) design our RL agents utilizing ‘pressure’ concept to achieve signal coordination in region-level; (2) show that implicit coordination could be achieved by individual control agents with well-crafted reward design thus reducing the dimensionality; and (3) conduct extensive experiments on multiple scenarios, including a real-world scenario with 2510 traffic lights in Manhattan, New York City 1 2.

Cooperation among the traffic signals enables vehicles to move through intersections more quickly. Conventional transportation approaches implement cooperation by pre-calculating the offsets between two intersections. Such pre-calculated offsets are not suitable for dynamic traffic environments. To enable cooperation of traffic signals, in this paper, we propose a model, CoLight, which uses graph attentional networks to facilitate communication. Specifically, for a target intersection in a network, CoLight can not only incorporate the temporal and spatial influences of neighboring intersections to the target intersection, but also build up index-free modeling of neighboring intersections. To the best of our knowledge, we are the first to use graph attentional networks in the setting of reinforcement learning for traffic signal control and to conduct experiments on the large-scale road network with hundreds of traffic signals. In experiments, we demonstrate that by learning the communication, the proposed model can achieve superior performance against the state-of-the-art methods.

Traffic signal control is essential for transportation efficiency in road networks. It has been a challenging problem because of the complexity in traffic dynamics. Conventional transportation research suffers from the incompetency to adapt to dynamic traffic situations. Recent studies propose to use reinforcement learning (RL) to search for more efficient traffic signal plans. However, most existing RL-based studies design the key elements - reward and state - in a heuristic way. This results in highly sensitive performances and a long learning process. To avoid the heuristic design of RL elements, we propose to connect RL with recent studies in transportation research. Our method is inspired by the state-of-the-art method max pressure (MP) in the transportation field. The reward design of our method is well supported by the theory in MP, which can be proved to be maximizing the throughput of the traffic network, i.e., minimizing the overall network travel time. We also show that our concise state representation can fully support the optimization of the proposed reward function. Through comprehensive experiments, we demonstrate that our method outperforms both conventional transportation approaches and existing learning-based methods.