AIBullisharXiv – CS AI · Mar 66/10
🧠Researchers introduce the What Is Missing (WIM) rating system for Large Language Models that uses natural-language feedback instead of numerical ratings to improve preference learning. WIM computes ratings by analyzing cosine similarity between model outputs and judge feedback embeddings, producing more interpretable and effective training signals with fewer ties than traditional rating methods.
AINeutralarXiv – CS AI · Mar 45/103
🧠Researchers have developed new methods to understand how Video Diffusion Transformers convert motion-related text descriptions into video content. The study introduces GramCol and Interpretable Motion-Attentive Maps (IMAP) to spatially and temporally localize motion concepts in AI-generated videos without requiring gradient calculations.
AIBullisharXiv – CS AI · Mar 45/102
🧠Researchers have developed Domain-aware Fourier Features (DaFFs) to enhance Physics-Informed Neural Networks (PINNs), achieving orders-of-magnitude lower errors and faster convergence. The approach incorporates domain-specific characteristics like geometry and boundary conditions while eliminating the need for explicit boundary condition loss terms, making PINNs more accurate, efficient, and interpretable.
AINeutralarXiv – CS AI · Mar 36/107
🧠Researchers propose a new gauge-theoretic framework for understanding superposition in large language models, replacing traditional single-dictionary approaches with local semantic charts. The method introduces three measurable obstructions to interpretability and demonstrates results on Llama 3.2 3B model with various datasets.
AIBullisharXiv – CS AI · Mar 36/106
🧠Researchers propose BiCAM, a new method for interpreting Vision Transformer (ViT) decisions that captures both positive and negative contributions to predictions. The approach improves explanation quality and enables adversarial example detection across multiple ViT variants without requiring model retraining.
AIBullisharXiv – CS AI · Mar 36/103
🧠Researchers propose Explanation-Guided Adversarial Training (EGAT), a framework that combines adversarial training with explainable AI to create more robust and interpretable deep neural networks. The method achieves 37% improvement in adversarial accuracy while producing semantically meaningful explanations with only 16% increase in training time.
AIBullisharXiv – CS AI · Mar 26/1019
🧠Researchers have developed EMO-R3, a new framework that enhances emotional reasoning capabilities in Multimodal Large Language Models through reflective reinforcement learning. The approach introduces structured emotional thinking and reflective rewards to improve interpretability and emotional intelligence in visual understanding tasks.
AIBullisharXiv – CS AI · Mar 26/1013
🧠Researchers have developed a new method to extract interpretable causal mechanisms from neural networks using structured pruning as a search technique. The approach reframes network pruning as finding approximate causal abstractions, yielding closed-form criteria for simplifying networks while maintaining their causal structure under interventions.
AIBullisharXiv – CS AI · Mar 26/1021
🧠Researchers propose a training-free solution to reduce hallucinations in multimodal AI models by rebalancing attention between perception and reasoning layers. The method achieves 4.2% improvement in reasoning accuracy with minimal computational overhead.
AIBullisharXiv – CS AI · Feb 276/106
🧠Researchers propose an Evaluation Agent framework to assess AI agent decision-making in AutoML pipelines, moving beyond outcome-focused metrics to evaluate intermediate decisions. The system can detect faulty decisions with 91.9% F1 score and reveals impacts ranging from -4.9% to +8.3% in final performance metrics.
AIBullisharXiv – CS AI · Feb 276/104
🧠Researchers decoded the internal representations of scGPT, a single-cell foundation model, revealing it organizes genes into interpretable biological coordinate systems rather than opaque features. The model encodes cellular organization patterns including protein localization, interaction networks, and regulatory relationships across its transformer layers.
AIBullisharXiv – CS AI · Feb 276/106
🧠Researchers introduce Temporal Sparse Autoencoders (T-SAEs), a new method that improves AI model interpretability by incorporating temporal structure of language through contrastive loss. The technique enables better separation of semantic from syntactic features and recovers smoother, more coherent semantic concepts without sacrificing reconstruction quality.
AINeutralGoogle DeepMind Blog · Dec 166/105
🧠Google has released Gemma Scope 2, providing open interpretability tools for understanding the behavior of language models across the entire Gemma 3 family. These tools are designed to help the AI safety community analyze and interpret complex language model behaviors.
AIBullishOpenAI News · Nov 136/107
🧠OpenAI is researching mechanistic interpretability through sparse neural network models to better understand AI reasoning processes. This approach aims to make AI systems more transparent and improve their safety and reliability.
AIBullishHugging Face Blog · Jul 316/106
🧠Google has released Gemma 2 2B, a smaller 2-billion parameter version of its open-source AI model, alongside ShieldGemma for safety filtering and Gemma Scope for model interpretability. These releases expand Google's Gemma family with more accessible and transparent AI tools for developers and researchers.
AIBullishOpenAI News · Apr 146/105
🧠OpenAI has launched Microscope, a visualization tool that provides detailed views of layers and neurons in eight vision AI models commonly used in interpretability research. The tool aims to help researchers better understand and analyze the internal features that develop within neural networks.
AINeutralarXiv – CS AI · Mar 174/10
🧠Researchers have developed SyMPLER, an explainable AI model for time series forecasting that uses dynamic piecewise-linear approximations to handle nonstationary environments. The model automatically determines when to add new local models based on prediction errors using Statistical Learning Theory, achieving comparable performance to black-box models while maintaining interpretability.
AINeutralarXiv – CS AI · Mar 175/10
🧠Researchers introduce Jacobian Scopes, a new gradient-based method for interpreting how individual tokens influence Large Language Model predictions. The technique uses perturbation theory and information geometry to reveal model biases, translation strategies, and learning mechanisms, with open-source implementations and an interactive demo available.
🏢 Hugging Face
AINeutralarXiv – CS AI · Mar 54/10
🧠Researchers introduced StructLens, a new analytical framework that uses maximum spanning trees to reveal global structural relationships between layers in language models, going beyond existing local token analysis methods. The approach shows different similarity patterns compared to traditional cosine similarity and proves effective for practical applications like layer pruning.
AINeutralarXiv – CS AI · Mar 54/10
🧠Researchers introduce PatchDecomp, a new neural network method for time series forecasting that achieves high accuracy while providing interpretable explanations. The method divides time series into patches and shows how each patch contributes to predictions, offering both quantitative and visual insights into forecasting decisions.
AINeutralarXiv – CS AI · Mar 54/10
🧠Researchers developed TPK, a trajectory prediction system for autonomous vehicles that integrates prior knowledge to make predictions more trustworthy and physically feasible. The system incorporates interaction and kinematic models for vehicles, pedestrians, and cyclists, improving interpretability while ensuring predictions adhere to physics.
AINeutralarXiv – CS AI · Mar 54/10
🧠Researchers introduce WeightLens and CircuitLens, two new methods for analyzing neural network interpretability that go beyond traditional activation-based approaches. These tools aim to provide more systematic and scalable analysis of neural network circuits by interpreting features directly from weights and capturing feature interactions.
AINeutralarXiv – CS AI · Mar 34/104
🧠Researchers have developed a new Explainable AI method that makes Wasserstein distances more interpretable by attributing distance calculations to specific data components like subgroups and features. The framework enables better analysis of dataset shifts and transport phenomena across diverse applications with high accuracy.
AINeutralarXiv – CS AI · Mar 34/105
🧠Researchers analyzed how Large Language Models access semantic memory using the Semantic Fluency Task, finding that LLMs exhibit similar memory foraging patterns to humans. The study reveals convergent and divergent search strategies in LLMs that mirror human cognitive behavior, potentially enabling better human-AI alignment or productive cognitive disalignment.
AINeutralarXiv – CS AI · Mar 34/107
🧠Researchers successfully applied a Concept Induction framework for neural network interpretability to the SUN2012 dataset, demonstrating the method's broader applicability beyond the original ADE20K dataset. The study assigns interpretable semantic labels to hidden neurons in CNNs and validates them through statistical testing and web-sourced images.
