Recent advancements in Optical Character Recognition (OCR) have been driven by transformer-based models. OCR systems are critical in numerous high-stakes domains, yet their vulnerability to adversarial attack remains largely uncharted territory, raising concerns about security and compliance with emerging AI regulations. In this work we present a novel framework to assess the resilience of Transformer-based OCR (TrOCR) models. We develop and assess algorithms for both targeted and untargeted attacks. For the untargeted case, we measure the Character Error Rate (CER), while for the targeted case we use the success ratio. We find that TrOCR is highly vulnerable to untargeted attacks and somewhat less vulnerable to targeted attacks. On a benchmark handwriting data set, untargeted attacks can cause a CER of more than 1 without being noticeable to the eye. With a similar perturbation size, targeted attacks can lead to success rates of around $25\%$ -- here we attacked single tokens, requiring TrOCR to output the tenth most likely token from a large vocabulary.

Although much work has been done on explainability in the computer vision and natural language processing (NLP) fields, there is still much work to be done to explain methods applied to time series as time series by nature can not be understood at first sight. In this paper, we present a Deep Neural Network (DNN) in a teacher-student architecture (distillation model) that offers interpretability in time-series classification tasks. The explainability of our approach is based on transforming the time series to 2D plots and applying image highlight methods (such as LIME and GradCam), making the predictions interpretable. At the same time, the proposed approach offers increased accuracy competing with the baseline model with the trade-off of increasing the training time.

Evaluating the accuracy of outputs generated by Large Language Models (LLMs) is especially important in the climate science and policy domain. We introduce the Expert Confidence in Climate Statements (ClimateX) dataset, a novel, curated, expert-labeled dataset consisting of 8094 climate statements collected from the latest Intergovernmental Panel on Climate Change (IPCC) reports, labeled with their associated confidence levels. Using this dataset, we show that recent LLMs can classify human expert confidence in climate-related statements, especially in a few-shot learning setting, but with limited (up to 47%) accuracy. Overall, models exhibit consistent and significant over-confidence on low and medium confidence statements. We highlight implications of our results for climate communication, LLMs evaluation strategies, and the use of LLMs in information retrieval systems.

The investigation reported in this document focuses on identifying systems with symmetries using equivariant autoregressive reservoir computers. General results in structured matrix approximation theory are presented, exploring a two-fold approach. Firstly, a comprehensive examination of generic symmetry-preserving nonlinear time delay embedding is conducted. This involves analyzing time series data sampled from an equivariant system under study. Secondly, sparse least-squares methods are applied to discern approximate representations of the output coupling matrices. These matrices play a pivotal role in determining the nonlinear autoregressive representation of an equivariant system. The structural characteristics of these matrices are dictated by the set of symmetries inherent in the system. The document outlines prototypical algorithms derived from the described techniques, offering insight into their practical applications. Emphasis is placed on their effectiveness in the identification and predictive simulation of equivariant nonlinear systems, regardless of whether such systems exhibit chaotic behavior.

We introduce HallusionBench, a comprehensive benchmark designed for the evaluation of image-context reasoning. This benchmark presents significant challenges to advanced large visual-language models (LVLMs), such as GPT-4V(Vision) and LLaVA-1.5, by emphasizing nuanced understanding and interpretation of visual data. The benchmark comprises 346 images paired with 1129 questions, all meticulously crafted by human experts. We introduce a novel structure for these visual questions designed to establish control groups. This structure enables us to conduct a quantitative analysis of the models' response tendencies, logical consistency, and various failure modes. In our evaluation on HallusionBench, we benchmarked 13 different models, highlighting a 31.42% question-pair accuracy achieved by the state-of-the-art GPT-4V. Notably, all other evaluated models achieve accuracy below 16%. Moreover, our analysis not only highlights the observed failure modes, including language hallucination and visual illusion, but also deepens an understanding of these pitfalls. Our comprehensive case studies within HallusionBench shed light on the challenges of hallucination and illusion in LVLMs. Based on these insights, we suggest potential pathways for their future improvement. The benchmark and codebase can be accessed at https://github.com/tianyi-lab/HallusionBench.

Stochastic gradient descent (SGD) and adaptive gradient methods, such as Adam and RMSProp, have been widely used in training deep neural networks. We empirically show that while the difference between the standard generalization performance of models trained using these methods is small, those trained using SGD exhibit far greater robustness under input perturbations. Notably, our investigation demonstrates the presence of irrelevant frequencies in natural datasets, where alterations do not affect models' generalization performance. However, models trained with adaptive methods show sensitivity to these changes, suggesting that their use of irrelevant frequencies can lead to solutions sensitive to perturbations. To better understand this difference, we study the learning dynamics of gradient descent (GD) and sign gradient descent (signGD) on a synthetic dataset that mirrors natural signals. With a three-dimensional input space, the models optimized with GD and signGD have standard risks close to zero but vary in their adversarial risks. Our result shows that linear models' robustness to $\ell_2$-norm bounded changes is inversely proportional to the model parameters' weight norm: a smaller weight norm implies better robustness. In the context of deep learning, our experiments show that SGD-trained neural networks have smaller Lipschitz constants, explaining the better robustness to input perturbations than those trained with adaptive gradient methods.

For solving linear inverse problems, particularly of the type that appears in tomographic imaging and compressive sensing, this paper develops two new approaches. The first approach is an iterative algorithm that minimizes a regularized least squares objective function where the regularization is based on a compound Gaussian prior distribution. The compound Gaussian prior subsumes many of the commonly used priors in image reconstruction, including those of sparsity-based approaches. The developed iterative algorithm gives rise to the paper's second new approach, which is a deep neural network that corresponds to an "unrolling" or "unfolding" of the iterative algorithm. Unrolled deep neural networks have interpretable layers and outperform standard deep learning methods. This paper includes a detailed computational theory that provides insight into the construction and performance of both algorithms. The conclusion is that both algorithms outperform other state-of-the-art approaches to tomographic image formation and compressive sensing, especially in the difficult regime of low training.

Unrestricted adversarial attacks typically manipulate the semantic content of an image (e.g., color or texture) to create adversarial examples that are both effective and photorealistic, demonstrating their ability to deceive human perception and deep neural networks with stealth and success. However, current works usually sacrifice unrestricted degrees and subjectively select some image content to guarantee the photorealism of unrestricted adversarial examples, which limits its attack performance. To ensure the photorealism of adversarial examples and boost attack performance, we propose a novel unrestricted attack framework called Content-based Unrestricted Adversarial Attack. By leveraging a low-dimensional manifold that represents natural images, we map the images onto the manifold and optimize them along its adversarial direction. Therefore, within this framework, we implement Adversarial Content Attack (ACA) based on Stable Diffusion and can generate high transferable unrestricted adversarial examples with various adversarial contents. Extensive experimentation and visualization demonstrate the efficacy of ACA, particularly in surpassing state-of-the-art attacks by an average of 13.3-50.4% and 16.8-48.0%

A voting method is Condorcet consistent if in any election in which one candidate is preferred by majorities to each of the other candidates, this candidate--the Condorcet winner--is the unique winner of the election. Condorcet consistent voting methods form an important class of methods in the theory of voting (see, e.g., Fishburn 1977; Brams and Fishburn 2002, 8; Zwicker 2016, 2.4; Pacuit 2019, 3.1.1). Although Condorcet methods are not currently used in government elections, they have been used by several private organizations (see Wikipedia contributors 2020b) and in over 30,000 polls through the Condorcet Internet Voting Service (https://civs.cs.cornell.edu). Recent initiatives in the U.S. to make available Instant Runoff Voting (Kambhampaty 2019), which uses the same ranked ballots needed for Condorcet methods, bring Condorcet methods closer to political application. Indeed, Eric Maskin and Amartya Sen have recently proposed the use of Condorcet methods in U.S. presidential primaries (Maskin and Sen 2016, 2017a,b). In the meantime, Condorcet methods continue to be used by committees, clubs, etc.

Large Language Models (LLMs) are notorious for blending fact with fiction and generating non-factual content, known as hallucinations. To tackle this challenge, we propose an interactive system that helps users obtain insights into the reliability of the generated text. Our approach is based on the idea that the self-consistency of multiple samples generated by the same LLM relates to its confidence in individual claims in the generated texts. Using this idea, we design RELIC, an interactive system that enables users to investigate and verify semantic-level variations in multiple long-form responses. This allows users to recognize potentially inaccurate information in the generated text and make necessary corrections. From a user study with ten participants, we demonstrate that our approach helps users better verify the reliability of the generated text. We further summarize the design implications and lessons learned from this research for inspiring future studies on reliable human-LLM interactions.