Analternativeapproach isto predict instead adistributionp(ˆy|x) = Φˆy(x) over possible values of the annotationy. Theannotators are shown a set of points sampled randomly and uniformly over one of predefined body parts of aperson inan image.

Despite their impressive capabilities, Large Language Models (LLMs) exhibit unwanted uncertainty, a phenomenon where a model changes a previously correct answer into an incorrect one when re-prompted. This behavior undermines trust and poses serious risks in high-stakes domains. In this work, we investigate the mechanisms that drive this phenomenon. We adapt the Needle-in-a-Haystack retrieval framework and integrate a Flip-style re-evaluation prompt to simulate realistic answer-flipping scenarios. We find that retrieval heads are not primarily responsible for avoiding uncertainty. Instead, we identify a small set of non-retrieval attention heads that disproportionately attend to misleading tokens in uncertain contexts. Masking these heads yields significant improvements, reducing flip behavior by up to 15% without introducing incoherence or overcorrection. However, when tested for downstream tasks, we observe trade-offs with flip behavior. Our findings contribute to the growing field of mechanistic interpretability and present a simple yet effective technique for mitigating uncertainty-driven failure modes in LLMs.

Neural Information Processing Systems http://nips.cc/

Large Language Models (LLMs) have the tendency to hallucinate, i.e., to sporadically generate false or fabricated information. This presents a major challenge, as hallucinations often appear highly convincing and users generally lack the tools to detect them. Uncertainty quantification (UQ) provides a framework for assessing the reliability of model outputs, aiding in the identification of potential hallucinations. In this work, we introduce pre-trained UQ heads: supervised auxiliary modules for LLMs that substantially enhance their ability to capture uncertainty compared to unsupervised UQ methods. Their strong performance stems from the powerful Transformer architecture in their design and informative features derived from LLM attention maps. Experimental evaluation shows that these heads are highly robust and achieve state-of-the-art performance in claim-level hallucination detection across both in-domain and out-of-domain prompts. Moreover, these modules demonstrate strong generalization to languages they were not explicitly trained on. We pre-train a collection of UQ heads for popular LLM series, including Mistral, Llama, and Gemma 2. We publicly release both the code and the pre-trained heads.

Accurate feature detection is fundamental for various computer vision tasks, including autonomous robotics, 3D reconstruction, medical imaging, and remote sensing. Despite advancements in enhancing the robustness of visual features, no existing method measures the utility of visual information before processing by specific feature-type algorithms. To address this gap, we introduce PIXER and the concept of "Featureness," which reflects the inherent interest and reliability of visual information for robust recognition, independent of any specific feature type. Leveraging a generalization on Bayesian learning, our approach quantifies both the probability and uncertainty of a pixel's contribution to robust visual utility in a single-shot process, avoiding costly operations such as Monte Carlo sampling and permitting customizable featureness definitions adaptable to a wide range of applications. We evaluate PIXER on visual odometry with featureness selectivity, achieving an average of 31% improvement in RMSE trajectory with 49% fewer features.

Neural networks (NNs) are increasingly used in always-on safety-critical applications deployed on hardware accelerators (NN-HAs) employing various memory technologies. Reliable continuous operation of NN is essential for safety-critical applications. During online operation, NNs are susceptible to single and multiple permanent and soft errors due to factors such as radiation, aging, and thermal effects. Explicit NN-HA testing methods cannot detect transient faults during inference, are unsuitable for always-on applications, and require extensive test vector generation and storage. Therefore, in this paper, we propose the \emph{uncertainty fingerprint} approach representing the online fault status of NN. Furthermore, we propose a dual head NN topology specifically designed to produce uncertainty fingerprints and the primary prediction of the NN in \emph{a single shot}. During the online operation, by matching the uncertainty fingerprint, we can concurrently self-test NNs with up to $100\%$ coverage with a low false positive rate while maintaining a similar performance of the primary task. Compared to existing works, memory overhead is reduced by up to $243.7$ MB, multiply and accumulate (MAC) operation is reduced by up to $10000\times$, and false-positive rates are reduced by up to $89\%$.

Accurate and efficient uncertainty estimation is crucial to build reliable Machine Learning (ML) models capable to provide calibrated uncertainty estimates, generalize and detect Out-Of-Distribution (OOD) datasets. To this end, Deterministic Uncertainty Methods (DUMs) is a promising model family capable to perform uncertainty estimation in a single forward pass. This work investigates important design choices in DUMs: (1) we show that training schemes decoupling the core architecture and the uncertainty head schemes can significantly improve uncertainty performances. Safety is critical to the adoption of deep learning in domains such as autonomous driving, medical diagnosis, or financial trading systems. A solution for this problem is to create reliable models capable to estimate the uncertainty of its own predictions. Different uncertainty types are divided in aleatoric uncertainty quantified by the inherited noise in the data, thus irreducible; epistemic uncertainty quantified by the modeling choice or lack of data, thus reducible; predictive uncertainty, a combination of aleatoric and epistemic (Gal, 2016). In practice, high quality uncertainty estimates must be calibrated and able to detect Out-Of-Distribution (OOD) data like anomalies while preserving good Out-Of-Distribution (OOD) generalization performances like on dataset shifts. Recently, a family of methods for uncertainty estimation named Deterministic Uncertainty Methods (DUMs) have emerged (Postels et al., 2022). Contrary to uncertainty methods such as Ensembles (Lakshminarayanan et al., 2017), MC Dropout (Gal & Ghahramani, 2016) or other Bayesian neural networks on weights (Blundell et al., 2015), which require multiple forward passes to make predictions, DUMs only require a single forward pass, thus making them significantly more computationally efficient.