Interactive Naming for Explaining Deep Neural Networks: A Formative Study
Explore this paper's citation graph
Summary
A user interface for "interactive naming," which allows a human annotator to manually cluster significant activation maps in a test set into meaningful groups called "visual concepts", is developed and found that a large fraction of the activation maps have recognizable visual concepts, and that there is significant agreement between the different annotators about their denotations.
- Type
- preprint
- Published
- 2018-12-18
- Cited by
- 11
- References
- 29
- Access
- Open access
- OpenAlex
- https://openalex.org/W2905111770
- Semantic Scholar
- https://api.semanticscholar.org/CorpusID:56169668
Keywords
Computer science, Classifier (UML), Artificial intelligence, Formative assessment, Set (abstract data type)
References
- On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation
- The Caltech-UCSD Birds-200-2011 Dataset
- Structured labeling for facilitating concept evolution in machine learning
- Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
- “Why Should I Trust You?”: Explaining the Predictions of Any Classifier
- Not Just a Black Box: Learning Important Features Through Propagating Activation Differences
- Setwise Comparison: Consistent, Scalable, Continuum Labels for Computer Vision
- Axiomatic Attribution for Deep Networks
- Interpretable Explanations of Black Boxes by Meaningful Perturbation
- Comparing Two Clusterings Using Matchings between Clusters of Clusters
- Network Dissection: Quantifying Interpretability of Deep Visual Representations
- Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
- Visual Explanations for Convolutional Neural Networks via Input Resampling
- Embedding Deep Networks into Visual Explanations
- Top-Down Neural Attention by Excitation Backprop
- Generating Visual Explanations
- Learning how to explain neural networks: PatternNet and PatternAttribution
- A Unified Approach to Interpreting Model Predictions
- Documenting evidence of a reuse of ‘“why should I trust you?”: explaining the predictions of any classifier’
- Introduction to Information Retrieval
Cited by
- Do ML Experts Discuss Explainability for AI Systems?
- Explainable Artificial Intelligence: a Systematic Review
- Debiasing Concept Bottleneck Models with Instrumental Variables
- Classification of Explainable Artificial Intelligence Methods through Their Output Formats
- From Heatmaps to Structural Explanations of Image Classifiers
- Increasing the Value of XAI for Users: A Psychological Perspective
- Understanding User Preferences in Explainable Artificial Intelligence: A Mapping Function Proposal
- Debiasing Concept Bottleneck Models with a Causal Analysis Technique
- Debiasing Concept-based Explanations with Causal Analysis
- Bridging the Gap: Providing Post-Hoc Symbolic Explanations for Sequential Decision-Making Problems with Inscrutable Representations
- Leveraging PDDL to Make Inscrutable Agents Interpretable: A Case for Post Hoc Symbolic Explanations for Sequential-Decision Making Problems
- B RIDGING THE G AP : P ROVIDING P OST -H OC S YMBOLIC E XPLANATIONS FOR S EQUENTIAL D ECISION -M AKING P ROBLEMS WITH I NSCRUTABLE R EPRESENTATIONS Finding Algorithm for Finding Missing Precondition
Related papers
- Human-Centered Tools for Coping with Imperfect Algorithms During Medical Decision-Making
- Towards a universal detector by mining concepts with small semantic gaps
- Annotator rationales for visual recognition
- Making use of Semantic Concept Detection for Modelling Human Preferences in Visual Summarization
- Choosing Basic-Level Concept Names Using Visual and Language Context
- Informative Visual Storytelling with Cross-modal Rules
- Direct Semantic Analysis for Social Image Classification