Knowledge from Uncertainty in Evidential Deep Learning

Cai Davies, Marc Roig Vilamala, Alun Preece, Federico Cerutti, Lance Kaplan, Supriyo Chakraborty PDF
Abstract: This work reveals an {\em evidential signal} that emerges from the uncertainty value in Evidential Deep Learning (EDL). EDL is one example of a class of uncertainty-aware deep learning approaches designed to provide confidence (or epistemic uncertainty) about the current test sample. In particular for computer vision and bidirectional encoder large language models, the `evidential signal' arising from the Dirichlet strength in EDL can, in some cases, discriminate between classes, which is particularly strong when using large language models. We hypothesise that the KL regularisation term causes EDL to couple aleatoric and epistemic uncertainty. In this paper, we empirically investigate the correlations between misclassification and evaluated uncertainty, and show that EDL's `evidential signal' is due to misclassification bias. We critically evaluate EDL with other Dirichlet-based approaches, namely Generative Evidential Neural Networks (EDL-GEN) and Prior Networks, and show theoretically and empirically the differences between these loss functions. We conclude that EDL's coupling of uncertainty arises from these differences due to the use (or lack) of out-of-distribution samples during training.

Uncertainty-aware Grounded Action Transformation towards Sim-to-Real Transfer for Traffic Signal Control

Longchao Da, Hao Mei, Romir Sharma, Hua Wei PDF
Abstract: Traffic signal control (TSC) is a complex and important task that affects the daily lives of millions of people. Reinforcement Learning (RL) has shown promising results in optimizing traffic signal control, but current RL-based TSC methods are mainly trained in simulation and suffer from the performance gap between simulation and the real world. In this paper, we propose a simulation-to-real-world (sim-to-real) transfer approach called UGAT, which transfers a learned policy trained from a simulated environment to a real-world environment by dynamically transforming actions in the simulation with uncertainty to mitigate the domain gap of transition dynamics. We evaluate our method on a simulated traffic environment and show that it significantly improves the performance of the transferred RL policy in the real world.

Building One-class Detector for Anything: Open-vocabulary Zero-shot OOD Detection Using Text-image Models

Yunhao Ge, Jie Ren, Jiaping Zhao, Kaifeng Chen, Andrew Gallagher, Laurent Itti, Balaji Lakshminarayanan PDF
Abstract: We focus on the challenge of out-of-distribution (OOD) detection in deep learning models, a crucial aspect in ensuring reliability. Despite considerable effort, the problem remains significantly challenging in deep learning models due to their propensity to output over-confident predictions for OOD inputs. We propose a novel one-class open-set OOD detector that leverages text-image pre-trained models in a zero-shot fashion and incorporates various descriptions of in-domain and OOD. Our approach is designed to detect anything not in-domain and offers the flexibility to detect a wide variety of OOD, defined via fine- or coarse-grained labels, or even in natural language. We evaluate our approach on challenging benchmarks including large-scale datasets containing fine-grained, semantically similar classes, distributionally shifted images, and multi-object images containing a mixture of in-domain and OOD objects. Our method shows superior performance over previous methods on all benchmarks.

Uncertainty Quantification of Deep Learning for Spatiotemporal Data: Challenges and Opportunities

Wenchong He, Zhe Jiang PDF
Abstract: With the advancement of GPS, remote sensing, and computational simulations, large amounts of geospatial and spatiotemporal data are being collected at an increasing speed. Such emerging spatiotemporal big data assets, together with the recent progress of deep learning technologies, provide unique opportunities to transform society. However, it is widely recognized that deep learning sometimes makes unexpected and incorrect predictions with unwarranted confidence, causing severe consequences in high-stake decision-making applications (e.g., disaster management, medical diagnosis, autonomous driving). Uncertainty quantification (UQ) aims to estimate a deep learning model's confidence. This paper provides a brief overview of UQ of deep learning for spatiotemporal data, including its unique challenges and existing methods. We particularly focus on the importance of uncertainty sources. We also identify several future research directions related to spatiotemporal data.

Delphic Offline Reinforcement Learning under Nonidentifiable Hidden Confounding

Alizée Pace, Hugo Yèche, Bernhard Schölkopf, Gunnar Ratsch, Tennenholtz Guy PDF
Abstract: A prominent challenge of offline reinforcement learning (RL) is the issue of hidden confounding: unobserved variables may influence both the actions taken by the agent and the observed outcomes. Hidden confounding can compromise the validity of any causal conclusion drawn from data and presents a major obstacle to effective offline RL. In the present paper, we tackle the problem of hidden confounding in the nonidentifiable setting. We propose a definition of uncertainty due to hidden confounding bias, termed delphic uncertainty, which uses variation over world models compatible with the observations, and differentiate it from the well-known epistemic and aleatoric uncertainties. We derive a practical method for estimating the three types of uncertainties, and construct a pessimistic offline RL algorithm to account for them. Our method does not assume identifiability of the unobserved confounders, and attempts to reduce the amount of confounding bias. We demonstrate through extensive experiments and ablations the efficacy of our approach on a sepsis management benchmark, as well as on electronic health records. Our results suggest that nonidentifiable hidden confounding bias can be mitigated to improve offline RL solutions in practice.

Evaluating Uncertainty Quantification for Bird’s Eye View Semantic Segmentation

Bowen Yang, Linlin Yu, Tianhao Wang, Changbin Li, Feng Chen PDF
Abstract: The fusion of raw features from multiple sensors (such as stereo cameras, LiDAR, radars) equipped on an autonomous vehicle to create a Bird's Eye View (BEV) representation is an essential and powerful component in planning and controlling systems. Recently, there has been a significant surge of interest in utilizing deep learning models for BEV semantic segmentation. However, being able to anticipate errors in segmentation and improve the explainability of DNNs is crucial to autonomous driving. In this paper, we evaluate various uncertainty quantification methods for BEV semantic segmentation based on two benchmark datasets (CARLA and nuScenes) with two representative backbones (Lift-Splat-Shoot and cross-view transformer). We perform an extensive evaluation of several uncertainty quantification methods. Among these methods, the evidential and postnet methods consistently demonstrate better performance in uncertainty quantification compared to MC dropout, ensemble, and deterministic baseline methods. Additionally, the ensemble method consistently exhibits the best performance in segmentation. We also propose augmenting uncertainty-aware BEV semantic segmentation models with supervised camera-view semantic segmentation features. Through extensive experiments, we consistently observe improvements in both the quality of BEV segmentation and the quality of uncertainty quantification. These findings suggest that exploring different types of supervision holds promise as a direction for enhancing uncertainty-aware BEV segmentation models.