Grokking LIME: How can we explain why an image classifier "knows" what’s in a photo...
summary
Uses MobileNet pre-trained on ImageNet to classify a 224×224 photo and isolates why class index 759 (reflex camera) scores highest. Segments image into about 94 superpixels, generates 128 perturbed images by blacking out random segments, runs the model on each, then fits a ridge linear model to get 94 coefficients marking segment importance. Provides a rendered heatmap visualizer; explanations fluctuate across runs so stability tuning and repeated sampling are necessary for reliable, user-meaningful output.
AI-written summary · about
chapters (14)
- 0:00Start
- 1:20Interpretable Machine Learning
- 1:49White Box Model
- 3:25The Four Principles of Explainable Ai That Have Been Published by Nist
- 5:07Knowledge Limits
- 7:19Image Classification Model
- 10:37Segment the Image
- 11:17Segmentation Functions
- 11:33Segment Mask
- 12:43The Segment Mask
- 19:39Approximate the Model in the Vicinity of the Original Image
- 20:05Calculate the Similarity between the Images
- 25:06What's the Main Difference between Lime and S-H -
- 29:03Do You Think the Explanations Could Also Help as an Input to Downstream Models
related talks
22mInterpreting Machine Learning Models with LIME
30mTowards a more transparent AI - Decrypting AI models using LIME
09mQuantitative Evaluation of Machine Learning Explanations: A Human-Grounded Benchmark
05mFine-tuned SigLIP Image
01h 05mEvaluating Model Explainability Methods
28mMem2Img: Memory-Resident Malware Detection via Convolution Neural Network
33mExplainable AI in the LIME-light
27mPattern Analysis: Learned vs. Engineered