MindCraft: How Concept Trees Take Shape In Deep Models

2025/09/26 by Bowei Tian, Tian, Bowei, Yexiao He +9
Computer Science · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Semantic Web and Ontologies

paper · pdf · doi:10.48550/arxiv.2510.03265

openalex publication_date 2025/09/26 · openalex created_date 2025/10/09 · openalex updated_date 2026/07/28

Abstract

Large-scale foundation models demonstrate strong performance across language, vision, and reasoning tasks. However, how they internally structure and stabilize concepts remains elusive. Inspired by causal inference, we introduce the MindCraft framework built upon Concept Trees. By applying spectral decomposition at each layer and linking principal directions into branching Concept Paths, Concept Trees reconstruct the hierarchical emergence of concepts, revealing exactly when they diverge from shared representations into linearly separable subspaces. Empirical evaluations across diverse scenarios across disciplines, including medical diagnosis, physics reasoning, and political decision-making, show that Concept Trees recover semantic hierarchies, disentangle latent concepts, and can be widely applied across multiple domains. The Concept Tree establishes a widely applicable and powerful framework that enables in-depth analysis of conceptual representations in deep models, marking a significant step forward in the foundation of interpretable AI.

Citations

• The Origins of Representation Manifolds in Large Language Models
• The Linear Representation Hypothesis and the Geometry of Large Language Models
• Representation Engineering: A Top-Down Approach to AI Transparency
• Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla
• DINOv2: Learning Robust Visual Features without Supervision
• LLaMA: Open and Efficient Foundation Language Models
• Scaling Vision Transformers to 22 Billion Parameters
• Locating and Editing Factual Associations in GPT
• Acquisition of Chess Knowledge in AlphaZero
• TruthfulQA: Measuring How Models Mimic Human Falsehoods
• Emerging Properties in Self-Supervised Vision Transformers
• SimCSE: Simple Contrastive Learning of Sentence Embeddings
• SimCSE: Simple Contrastive Learning of Sentence Embeddings
• BERT has a Moral Compass: Improvements of ethical and moral values of machines
• Similarity of Neural Network Representations Revisited
• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• On the importance of single directions for generalization
• Learning to Generate Reviews and Discovering Sentiment
• Towards A Rigorous Science of Interpretable Machine Learning
• The Mythos of Model Interpretability
• "Why Should I Trust You?": Explaining the Predictions of Any Classifier
• Learning Deep Features for Discriminative Localization
• Deep Residual Learning for Image Recognition
• Striving for Simplicity: The All Convolutional Net
• Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
• Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
• Visualizing and Understanding Convolutional Networks
• Distributed Representations of Words and Phrases and their Compositionality
• Speech Recognition with Deep Recurrent Neural Networks
• ImageNet classification with deep convolutional neural networks
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Related

• Learning Interpretable Concepts: Unifying Causal Representation Learning and Foundation Models
• Simple Mechanisms for Representing, Indexing and Manipulating Concepts
• ConceptFlow: Hierarchical and Fine-grained Concept-Based Explanation for Convolutional Neural Networks
• Concept-Based Mechanistic Interpretability Using Structured Knowledge Graphs
• Conceptual Modeling and Artificial Intelligence: A Systematic Mapping Study