Concepts from Representations: Post-hoc Concept Bottleneck Models via Sparse Decomposition of Visual Representations

Deep learning has achieved remarkable success in image recognition, yet their inherent opacity poses challenges for deployment in critical domains. Concept-based interpretations aim to address this by explaining model reasoning through human-understandable concepts. However, existing post-hoc methods and ante-hoc concept bottleneck models (CBMs), suffer from limitations such as unreliable concept relevance, non-visual or labor-intensive concept definitions, and model or data-agnostic assumptions. This paper introduces Post-hoc Concept Bottleneck Model via Representation Decomposition (PCBM-ReD), a novel pipeline that retrofits interpretability onto pretrained opaque models. PCBM-ReD automatically extracts visual concepts from a pre-trained encoder, employs multimodal large language models (MLLMs) to label and filter concepts based on visual identifiability and task relevance, and selects an independent subset via reconstruction-guided optimization. Leveraging CLIP's visual-text alignment, it decomposes image representations into linear combination of concept embeddings to fit into the CBMs abstraction. Extensive experiments across 11 image classification tasks show PCBM-ReD achieves state-of-the-art accuracy, narrows the performance gap with end-to-end models, and exhibits better interpretability.

Legal experts disagree with rationale extraction techniques for explaining ECtHR case outcome classification

Interpretability is critical for applications of large language models in the legal domain which requires trust and transparency. While some studies develop task-specific approaches, other use the classification model's parameters to explain the decisions. However, which technique explains the legal outcome prediction best remains an open question. To address this challenge, we propose a comparative analysis framework for model-agnostic interpretability techniques. Among these, we employ two rationale extraction methods, which justify outcomes with human-interpretable and concise text fragments (i.e., rationales) from the given input text. We conduct comparison by evaluating faithfulness-via normalized sufficiency and comprehensiveness metrics along with plausibility-by asking legal experts to evaluate extracted rationales. We further assess the feasibility of LLM-as-a-Judge using legal expert evaluation results. We show that the model's "reasons" for predicting a violation differ substantially from those of legal experts, despite highly promising quantitative analysis results and reasonable downstream classification performance. The source code of our experiments is publicly available at https://github.com/trusthlt/IntEval.

Soft Contextualized Encoder For User Defined Text Classification

User-Defined Text Classification (UDTC) considers the challenge of classifying input text to user-specified, previously unseen classes, a setting that arises frequently in real-world applications such as enterprise analytics, content moderation, and domain-specific information retrieval. We propose a soft-contextualized encoder architecture for UDTC which contextualizes each candidate label with the label set and a static soft prompt representation of the input query. Training on diverse, multi-source datasets enables the model to generalize effectively to zero-shot classification over entirely unseen topic sets drawn from arbitrary domains. We evaluate the proposed architecture both on held-out in-distribution test data and on multiple unseen UDTC benchmarks. Across datasets, the model achieves state-of-the-art performance, consistently outperforming or matching the baselines.

PDFInspect: A Unified Feature Extraction Framework for Malicious Document Detection

The increasing prevalence of malicious Portable Document Format (PDF) files necessitates robust and comprehensive feature extraction techniques for effective detection and analysis. This work presents a unified framework that integrates graph-based, structural, and metadata-driven analysis to generate a rich feature representation for each PDF document. The system extracts text from PDF pages and constructs undirected graphs based on pairwise word relationships, enabling the computation of graph-theoretic features such as node count, edge density, and clustering coefficient. Simultaneously, the framework parses embedded metadata to quantify character distributions, entropy patterns, and inconsistencies across fields such as author, title, and producer. Temporal features are derived from creation and modification timestamps to capture behavioral signatures, while structural elements including, object streams, fonts, and embedded images, are quantified to reflect document complexity. Boolean flags for potentially malicious PDF constructs (e.g., JavaScript, launch actions) are also extracted. Together, these features form a high-dimensional vector representation (170 dimensions) that is well-suited for downstream tasks such as malware classification, anomaly detection, and forensic analysis. The proposed approach is scalable, extensible, and designed to support real-world PDF threat intelligence workflows.6

AI Co-Scientist for Knowledge Synthesis in Medical Contexts: A Proof of Concept

Research waste in biomedical science is driven by redundant studies, incomplete reporting, and the limited scalability of traditional evidence synthesis workflows. We present an AI co-scientist for scalable and transparent knowledge synthesis based on explicit formalization of Population, Intervention, Comparator, Outcome, and Study design (PICOS). The platform integrates relational storage, vector-based semantic retrieval, and a Neo4j knowledge graph. Evaluation was conducted on dementia-sport and non-communicable disease corpora. Automated PICOS compliance and study design classification from titles and abstracts were performed using a Bidirectional Long Short-Term Memory baseline and a transformer-based multi-task classifier fine-tuned from PubMedBERT. Full-text synthesis employed retrieval-augmented generation with hybrid vector and graph retrieval, while BERTopic was used to identify thematic structure, redundancy, and evidence gaps. The transformer model achieved 95.7% accuracy for study design classification with strong agreement against expert annotations, while the Bi-LSTM achieved 87% accuracy for PICOS compliance detection. Retrieval-augmented generation outperformed non-retrieval generation for queries requiring structured constraints, cross-study integration, and graph-based reasoning, whereas non-retrieval approaches remained competitive for high-level summaries. Topic modeling revealed substantial thematic redundancy and identified underexplored research areas. These results demonstrate that PICOS-aware and explainable natural language processing can improve the scalability, transparency, and efficiency of evidence synthesis. The proposed architecture is domain-agnostic and offers a practical framework for reducing research waste across biomedical disciplines.

Fusion Matters: Length-Aware Analysis of Positional-Encoding Fusion in Transformers

Transformers require positional encodings to represent sequence order, yet most prior work focuses on designing new positional encodings rather than examining how positional information is fused with token embeddings. In this paper, we study whether the fusion mechanism itself affects performance, particularly in long-sequence settings. We conduct a controlled empirical study comparing three canonical fusion strategies--element-wise addition, concatenation with projection, and scalar gated fusion--under identical Transformer architectures, data splits, and random seeds. Experiments on three text classification datasets spanning short (AG News), medium (IMDB), and long (ArXiv) sequences show that fusion choice has negligible impact on short texts but produces consistent gains on long documents. To verify that these gains are structural rather than stochastic, we perform paired-seed analysis and cross-dataset comparison across sequence-length regimes. Additional experiments on the ArXiv dataset indicate that the benefit of learnable fusion generalizes across multiple positional encoding families. Finally, we explore a lightweight convolutional gating mechanism that introduces local inductive bias at the fusion level, evaluated on long documents only. Our results indicate that positional-encoding fusion is a non-trivial design choice for long-sequence Transformers and should be treated as an explicit modeling decision rather than a fixed default.

CAOS: Conformal Aggregation of One-Shot Predictors

One-shot prediction enables rapid adaptation of pretrained foundation models to new tasks using only one labeled example, but lacks principled uncertainty quantification. While conformal prediction provides finite-sample coverage guarantees, standard split conformal methods are inefficient in the one-shot setting due to data splitting and reliance on a single predictor. We propose Conformal Aggregation of One-Shot Predictors (CAOS), a conformal framework that adaptively aggregates multiple one-shot predictors and uses a leave-one-out calibration scheme to fully exploit scarce labeled data. Despite violating classical exchangeability assumptions, we prove that CAOS achieves valid marginal coverage using a monotonicity-based argument. Experiments on one-shot facial landmarking and RAFT text classification tasks show that CAOS produces substantially smaller prediction sets than split conformal baselines while maintaining reliable coverage.

Zero-Shot Distracted Driver Detection via Vision Language Models with Double Decoupling

Distracted driving is a major cause of traffic collisions, calling for robust and scalable detection methods. Vision-language models (VLMs) enable strong zero-shot image classification, but existing VLM-based distracted driver detectors often underperform in real-world conditions. We identify subject-specific appearance variations (e.g., clothing, age, and gender) as a key bottleneck: VLMs entangle these factors with behavior cues, leading to decisions driven by who the driver is rather than what the driver is doing. To address this, we propose a subject decoupling framework that extracts a driver appearance embedding and removes its influence from the image embedding prior to zero-shot classification, thereby emphasizing distraction-relevant evidence. We further orthogonalize text embeddings via metric projection onto Stiefel manifold to improve separability while staying close to the original semantics. Experiments demonstrate consistent gains over prior baselines, indicating the promise of our approach for practical road-safety applications.

CytoCLIP: Learning Cytoarchitectural Characteristics in Developing Human Brain Using Contrastive Language Image Pre-Training

The functions of different regions of the human brain are closely linked to their distinct cytoarchitecture, which is defined by the spatial arrangement and morphology of the cells. Identifying brain regions by their cytoarchitecture enables various scientific analyses of the brain. However, delineating these areas manually in brain histological sections is time-consuming and requires specialized knowledge. An automated approach is necessary to minimize the effort needed from human experts. To address this, we propose CytoCLIP, a suite of vision-language models derived from pre-trained Contrastive Language-Image Pre-Training (CLIP) frameworks to learn joint visual-text representations of brain cytoarchitecture. CytoCLIP comprises two model variants: one is trained using low-resolution whole-region images to understand the overall cytoarchitectural pattern of an area, and the other is trained on high-resolution image tiles for detailed cellular-level representation. The training dataset is created from NISSL-stained histological sections of developing fetal brains of different gestational weeks. It includes 86 distinct regions for low-resolution images and 384 brain regions for high-resolution tiles. We evaluate the model's understanding of the cytoarchitecture and generalization ability using region classification and cross-modal retrieval tasks. Multiple experiments are performed under various data setups, including data from samples of different ages and sectioning planes. Experimental results demonstrate that CytoCLIP outperforms existing methods. It achieves an F1 score of 0.87 for whole-region classification and 0.91 for high-resolution image tile classification.

Predicting When to Trust Vision-Language Models for Spatial Reasoning

Vision-Language Models (VLMs) demonstrate impressive capabilities across multimodal tasks, yet exhibit systematic spatial reasoning failures, achieving only 49% (CLIP) to 54% (BLIP-2) accuracy on basic directional relationships. For safe deployment in robotics and autonomous systems, we need to predict when to trust VLM spatial predictions rather than accepting all outputs. We propose a vision-based confidence estimation framework that validates VLM predictions through independent geometric verification using object detection. Unlike text-based approaches relying on self-assessment, our method fuses four signals via gradient boosting: geometric alignment between VLM claims and coordinates, spatial ambiguity from overlap, detection quality, and VLM internal uncertainty. We achieve 0.674 AUROC on BLIP-2 (34.0% improvement over text-based baselines) and 0.583 AUROC on CLIP (16.1% improvement), generalizing across generative and classification architectures. Our framework enables selective prediction: at 60% target accuracy, we achieve 61.9% coverage versus 27.6% baseline (2.2x improvement) on BLIP-2. Feature analysis reveals vision-based signals contribute 87.4% of model importance versus 12.7% from VLM confidence, validating that external geometric verification outperforms self-assessment. We demonstrate reliable scene graph construction where confidence-based pruning improves precision from 52.1% to 78.3% while retaining 68.2% of edges.