Page Stream Segmentation (PSS) is an essential prerequisite for automated document processing at scale. However, research progress has been limited by the absence of realistic public benchmarks. This paper works towards addressing this gap by introducing TABME++, an enhanced benchmark featuring commercial Optical Character Recognition (OCR) annotations. We evaluate the performance of large language models (LLMs) on PSS, focusing on decoder-based models fine-tuned with parameter-efficient methods. Our results show that decoder-based LLMs outperform smaller multimodal encoders. Through a review of existing PSS research and datasets, we identify key challenges and advancements in the field. Our findings highlight the key importance of robust OCR, providing valuable insights for the development of more effective document processing systems.

The rapid advancement of Large Language Models (LLMs) has significantly influenced various domains, leveraging their exceptional few-shot and zero-shot learning capabilities. In this work, we aim to explore and understand the LLMs-based feature selection methods from a data-centric perspective. We begin by categorizing existing feature selection methods with LLMs into two groups: data-driven feature selection which requires samples values to do statistical inference and text-based feature selection which utilizes prior knowledge of LLMs to do semantical associations using descriptive context. We conduct extensive experiments in both classification and regression tasks with LLMs in various sizes (e.g., GPT-4, ChatGPT and LLaMA-2). Our findings emphasize the effectiveness and robustness of text-based feature selection methods and showcase their potentials using a real-world medical application. We also discuss the challenges and future opportunities in employing LLMs for feature selection, offering insights for further research and development in this emerging field.

Anomaly detection is fundamental yet, challenging problem with practical applications in industry. The current approaches neglect the higher-order dependencies within the networks of interconnected sensors in the high-dimensional time series(multisensor data) for anomaly detection. To this end, we present a self-adapting anomaly detection framework for joint learning of (a) discrete hypergraph structure and (b) modeling the temporal trends and spatial relations among the interdependent sensors using the hierarchical encoder-decoder architecture to overcome the challenges. The hypergraph representation learning-based framework exploits the relational inductive biases in the hypergraph-structured data to learn the pointwise single-step-ahead forecasts through the self-supervised autoregressive task and predicts the anomalies based on the forecast error. Furthermore, our framework incentivizes learning the anomaly-diagnosis ontology through a differentiable approach. It derives the anomaly information propagation-based computational hypergraphs for root cause analysis and provides recommendations through an offline, optimal predictive control policy to remedy an anomaly. We conduct extensive experiments to evaluate the proposed method on the benchmark datasets for fair and rigorous comparison with the popular baselines. The proposed method outperforms the baseline models and achieves SOTA performance. We report the ablation studies to support the efficacy of the framework.

$K$-nearest neighbor language models ($k$NN-LMs), which integrate retrieval with next-word prediction, have demonstrated strong performance in language modeling as well as downstream NLP benchmarks. These results have led researchers to argue that models trained on poor quality or outdated data could perform well by employing a $k$NN extension that has access to a higher-quality datastore. In this work, we ask whether this improved ability to recall information really translates into downstream abilities. We extensively evaluate $k$NN-LMs on a diverse set of tasks, ranging from sentiment classification and commonsense reasoning to multi-hop reasoning. Results show that $k$NN-LMs excel at memory-intensive tasks, where utilizing the patterns in the input is sufficient for determining the output, but struggle with reasoning tasks that require integrating multiple pieces of information to derive new knowledge. We further demonstrate through oracle experiments and qualitative analysis that even with perfect retrieval, $k$NN-LMs still fail to determine the correct answers, placing an upper bound on their reasoning performance. Code and datastores are released at https://github.com/GSYfate/knnlm-limits/.

Brain tumors in magnetic resonance imaging (MR) are difficult, time-consuming, and prone to human error. These challenges can be resolved by developing automatic brain tumor segmentation methods from MR images. Various deep-learning models based on the U-Net have been proposed for the task. These deep-learning models are trained on a dataset of tumor images and then used for segmenting the masks. Mini-batch training is a widely used method in deep learning for training. However, one of the significant challenges associated with this approach is that if the training dataset has under-represented samples or samples with complex latent representations, the model may not generalize well to these samples. The issue leads to skewed learning of the data, where the model learns to fit towards the majority representations while underestimating the under-represented samples. The proposed dynamic batch training method addresses the challenges posed by under-represented data points, data points with complex latent representation, and imbalances within the class, where some samples may be harder to learn than others. Poor performance of such samples can be identified only after the completion of the training, leading to the wastage of computational resources. Also, training easy samples after each epoch is an inefficient utilization of computation resources. To overcome these challenges, the proposed method identifies hard samples and trains such samples for more iterations compared to easier samples on the BraTS2020 dataset. Additionally, the samples trained multiple times are identified and it provides a way to identify hard samples in the BraTS2020 dataset. The comparison of the proposed training approach with U-Net and other models in the literature highlights the capabilities of the proposed training approach.

Understanding biological processes, drug development, and biotechnological advancements requires detailed analysis of protein structures and sequences, a task in protein research that is inherently complex and time-consuming when performed manually. To streamline this process, we introduce ProteinGPT, a state-of-the-art multi-modal protein chat system, that allows users to upload protein sequences and/or structures for comprehensive protein analysis and responsive inquiries. ProteinGPT seamlessly integrates protein sequence and structure encoders with linear projection layers for precise representation adaptation, coupled with a large language model (LLM) to generate accurate and contextually relevant responses. To train ProteinGPT, we construct a large-scale dataset of 132,092 proteins with annotations, and optimize the instruction-tuning process using GPT-4o. This innovative system ensures accurate alignment between the user-uploaded data and prompts, simplifying protein analysis. Experiments show that ProteinGPT can produce promising responses to proteins and their corresponding questions.

Dual-encoder-based audio retrieval systems are commonly optimized with contrastive learning on a set of matching and mismatching audio-caption pairs. This leads to a shared embedding space in which corresponding items from the two modalities end up close together. Since audio-caption datasets typically only contain matching pairs of recordings and descriptions, it has become common practice to create mismatching pairs by pairing the audio with a caption randomly drawn from the dataset. This is not ideal because the randomly sampled caption could, just by chance, partly or entirely describe the audio recording. However, correspondence information for all possible pairs is costly to annotate and thus typically unavailable; we, therefore, suggest substituting it with estimated correspondences. To this end, we propose a two-staged training procedure in which multiple retrieval models are first trained as usual, i.e., without estimated correspondences. In the second stage, the audio-caption correspondences predicted by these models then serve as prediction targets. We evaluate our method on the ClothoV2 and the AudioCaps benchmark and show that it improves retrieval performance, even in a restricting self-distillation setting where a single model generates and then learns from the estimated correspondences. We further show that our method outperforms the current state of the art by 1.6 pp. mAP@10 on the ClothoV2 benchmark.

This study addresses the challenge of fleet design optimization in the context of heterogeneous multi-robot fleets, aiming to obtain feasible designs that balance performance and costs. In the domain of autonomous multi-robot exploration, reinforcement learning agents play a central role, offering adaptability to complex terrains and facilitating collaboration among robots. However, modifying the fleet composition results in changes in the learned behavior, and training multi-robot systems using multi-agent reinforcement learning is expensive. Therefore, an exhaustive evaluation of each potential fleet design is infeasible. To tackle these hurdles, we introduce Bayesian Optimization for Fleet Design (BOFD), a framework leveraging multi-objective Bayesian Optimization to explore fleets on the Pareto front of performance and cost while accounting for uncertainty in the design space. Moreover, we establish a sub-linear bound for cumulative regret, supporting BOFD's robustness and efficacy. Extensive benchmark experiments in synthetic and simulated environments demonstrate the superiority of our framework over state-of-the-art methods, achieving efficient fleet designs with minimal fleet evaluations.

How do people understand and evaluate claims about others' beliefs, even though these beliefs cannot be directly observed? In this paper, we introduce a cognitive model of epistemic language interpretation, grounded in Bayesian inferences about other agents' goals, beliefs, and intentions: a language-augmented Bayesian theory-of-mind (LaBToM). By translating natural language into an epistemic ``language-of-thought'', then evaluating these translations against the inferences produced by inverting a probabilistic generative model of rational action and perception, LaBToM captures graded plausibility judgments about epistemic claims. We validate our model in an experiment where participants watch an agent navigate a maze to find keys hidden in boxes needed to reach their goal, then rate sentences about the agent's beliefs. In contrast with multimodal LLMs (GPT-4o, Gemini Pro) and ablated models, our model correlates highly with human judgments for a wide range of expressions, including modal language, uncertainty expressions, knowledge claims, likelihood comparisons, and attributions of false belief.

Existing symbolic regression methods organize the space of candidate mathematical expressions primarily based on their syntactic, structural similarity. However, this approach overlooks crucial equivalences between expressions that arise from mathematical symmetries, such as commutativity, associativity, and distribution laws for arithmetic operations. Consequently, expressions with similar errors on a given data set are apart from each other in the search space. This leads to a rough error landscape in the search space that efficient local, gradient-based methods cannot explore. This paper proposes and implements a measure of a behavioral distance, BED, that clusters together expressions with similar errors. The experimental results show that the stochastic method for calculating BED achieves consistency with a modest number of sampled values for evaluating the expressions. This leads to computational efficiency comparable to the tree-based syntactic distance. Our findings also reveal that BED significantly improves the smoothness of the error landscape in the search space for symbolic regression.