AINeutralarXiv – CS AI · Jun 256/10
🧠Researchers developed an interpretable AI pipeline to analyze climate discourse across paid Meta advertisements and organic Bluesky posts from mid-2024 to mid-2025, revealing fundamental differences in messaging: paid platforms emphasize solution promotion in formal tones, while public social media centers on systemic critique with scientific grounding. The framework demonstrates how LLM-powered thematic analysis can surface structural differences in communication across heterogeneous platforms.
AINeutralarXiv – CS AI · Jun 256/10
🧠Researchers present Expresso-AI, a framework for interpreting deep learning models trained on facial videos to diagnose depression severity. The approach combines explainability with improved predictive performance by analyzing facial regions and temporal expression patterns, addressing a critical gap in automated mental health diagnosis where current methods lack interpretability.
AINeutralarXiv – CS AI · Jun 256/10
🧠Researchers demonstrate that logit distance—a measure based on differences in model predictions—better bounds representational similarity in neural networks than KL divergence does. The findings reveal that KL-based distillation can preserve predictive accuracy while failing to maintain the linear structure of internal representations, with implications for transfer learning and model compression.
AINeutralarXiv – CS AI · Jun 256/10
🧠Researchers identify 'cliff tokens'—specific points in LLM reasoning where a single token triggers failure in mathematical problem-solving. By deleting these tokens and resampling, models recover near-perfect accuracy, demonstrating that failures stem from precise decision points rather than diffuse errors. A taxonomy of cliff types enables targeted optimization that improves model reasoning by up to 6.6%.
AINeutralarXiv – CS AI · Jun 255/10
🧠Researchers analyzed how a Wav2Vec 2.0-based machine learning model interprets acoustic features in speech from oral and oropharyngeal cancer patients. Using canonical correlation analysis, they found the model's learned representations most strongly correlate with spectral and prosodic features, providing practical insights for improving pathological speech assessment systems.
AINeutralarXiv – CS AI · Jun 256/10
🧠Researchers introduce Joint Sparse Autoencoders (JSAE), a technique that improves how vision-language models can be analyzed and controlled by aligning visual and textual representations into shared, interpretable features. Testing across multiple VLM architectures reveals that steering interventions work most effectively at mid-to-late layers, offering insights for more precise multimodal model control.
🧠 Llama
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers propose CAFM, a Cohort-Anchored Foundation Model framework designed to improve interpretability and clinical reliability of AI systems for electronic health records by elevating patient cohorts to a primary learning object. The four-stage framework addresses limitations in existing EHR models through better data curation, cohort-conditioned training, multimodal alignment, and clinician feedback, with case studies demonstrating applications across kidney injury prediction, cardiovascular risk assessment, and imaging analysis.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers developed an Explainable AI framework using Federated Learning to identify career-related depression and anxiety among university students while preserving privacy. The model achieved 92.08% accuracy by analyzing behavioral data and facial expressions, successfully identifying key depression indicators consistent with psychological theory.
AINeutralarXiv – CS AI · Jun 236/10
🧠TraceView is an interactive visualization tool that helps developers understand and diagnose how LLM-based automated program repair agents work through their reasoning processes. By organizing agent trajectories into visual graphs with labeled components, the tool addresses a critical gap in debugging agent failures and improving repair outcomes.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers introduce Diffusion Integrated Gradients (DiffIG), a novel explainable AI method that uses diffusion models to generate optimized attribution paths instead of relying on fixed hand-crafted paths. The approach enables inference-time controllable feature attribution with improved explanation quality and perceptual alignment compared to existing path-based methods.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers introduce ForEx, a framework that translates LLM-generated explanations into formal logic (Lean4) to verify whether reasoning actually supports predicted labels on logical fallacy detection tasks. The study reveals a critical gap: while 90% of LLM outputs can be formally verified as logically sound, agreement with human annotations remains around 20%, exposing that formal correctness differs fundamentally from label accuracy.
AINeutralarXiv – CS AI · Jun 236/10
🧠A new arXiv paper argues that Large Language Models learn causal structure through a difference-making logic called variational induction, rather than through traditional causal inference frameworks like Pearl's interventionism. The research analyzes how LLM architectural features like token embeddings and self-attention implement this logic by identifying which word variations influence text predictions.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers propose a new method called Modified RISE-eval to evaluate attention map visualizations in AI speaker recognition systems. The study systematically reviews existing Class Activation Map (CAM)-based evaluation techniques and demonstrates how GradCAM and LayerCAM perform differently under various conditions, advancing the field of explainable AI (XAI) by making neural network decision-making more transparent and interpretable.
AIBullisharXiv – CS AI · Jun 236/10
🧠Researchers present IRR-Drive, an adaptive multimodal reflection framework that enhances autonomous driving systems by having Vision-Language-Action models explicitly reason about future consequences before generating trajectories. The system uses dual-modality reflection combining textual intentions with predicted bird's-eye view representations to self-correct decisions based on scene complexity, achieving state-of-the-art results on the NAVSIM benchmark.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers propose a novel Gaussian Process-based framework for medical image segmentation that explicitly models annotation bias and variability across multiple raters rather than encoding them implicitly. The approach improves uncertainty calibration in probabilistic predictions while maintaining segmentation accuracy, with quantifiable parameters reflecting individual annotator behavior.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers introduce LISDD, a framework for identifying where and why physics-based models fail by localizing errors to specific operating regimes and discovering sparse symbolic corrections. The method outperforms existing global-correction approaches by keeping parameter bias near zero while maintaining statistical rigor through finite-sample testing.
AINeutralarXiv – CS AI · Jun 236/10
🧠ReasoningLens, an open-source framework, addresses the transparency challenge posed by Large Reasoning Models' exceptionally long Chain-of-Thought traces. The tool provides hierarchical visualization, automated error detection, and diagnostic profiling to help researchers and developers interpret and optimize complex AI reasoning processes.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers identify 'premature commitment' as a hidden failure mode in LLM agents where models settle on an initial interpretation and defend it rather than adapting to new evidence. Using hidden-state analysis, they develop diagnostics that detect trajectory inconsistency with up to 97% accuracy and demonstrate that commitment is orthogonal to correctness—agents can be confidently wrong or right.
🧠 Llama
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers used sparse autoencoders to mechanistically analyze MolFormer, a chemical language model, revealing that it learns meaningful molecular semantics beyond surface-level syntax. Early layers track molecular grammar through position-encoding, while deeper layers capture pharmacologically relevant atomic features, with non-canonical SMILES notations causing more disruption than invalid ones due to cascading positional errors.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers propose NOMTO, a framework combining neural operators with symbolic equation discovery to identify governing equations from complex data involving nonlocal operators and memory effects. This advancement extends traditional symbolic discovery methods beyond local derivatives, enabling discovery of more realistic physical and mathematical models.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers have developed explainable AI techniques to improve trust and understanding of automatic speech recognition (ASR) systems by identifying minimal subsets of audio frames that cause specific transcriptions. The study adapts established XAI methods from image classification and evaluates them against multiple ASR systems including Google API and DeepSpeech using 100 audio samples.
AIBearisharXiv – CS AI · Jun 236/10
🧠Researchers developed a Shapley-value-based framework to quantify how adjectives steer Large Language Model outputs across architectures (GPT-4o-mini, Llama-3-70b, DeepSeek-R1, Phi-3, o3). The study reveals that steering effects are model-dependent, non-universal, and exhibit complex interaction patterns—larger models show unpredictable compositional behavior while smaller models respond more literally, challenging the viability of one-size-fits-all prompting strategies.
🧠 GPT-4
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers find that cross-attention mechanisms in speech-to-text models only explain about 50% of how the decoder attends to input, contradicting widespread assumptions that attention scores reliably indicate which parts of the audio are most relevant. The study across multiple model scales shows attention provides an incomplete view of the factors driving predictions.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers introduce CQD-SHAP, a framework that explains how neural models answer complex queries over incomplete knowledge graphs by computing the contribution of each query component using Shapley values from game theory. This approach addresses the black-box nature of existing complex query answering methods and demonstrates consistent effectiveness across multiple datasets.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers propose a Mamba-based architecture for multimodal medical data fusion that combines visual and tabular processing to improve cancer classification interpretability. Testing on skin and oral cancer datasets shows competitive performance with enhanced explainability through SHAP analysis, positioning state space models as viable alternatives to Transformers in medical AI applications.
