mlsanityData to 14 Oct 2026

qZACH-ViT: Quantization-Aware Intrinsic Explanations with Recursive Attribution-Stabilized Optimization

arXiv 2607.15421 ↗First posted 16 Jul 2026cs.LG cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Compact medical-image classifiers need efficiency and interpretable evidence, yet these goals are often addressed separately. We introduce qZACH-ViT, a quantization-aware extension of the zero-token (CLS-token-free), position-free ZACH-ViT backbone with recursive intrinsic patch-level class evidence. We also introduce Recursive Attribution-Stabilized Optimization (RASO), which norm-matches classification and attribution gradients and removes attribution components that conflict with classification. We evaluate four controlled conditions on seven MedMNIST datasets using 50 training images per class and ten fixed seeds, completing 280 runs. All 210 qZACH-ViT checkpoints are converted to executable mixed-precision ONNX INT8 graphs containing 16 signed INT8 MatMulInteger projections with INT32 accumulation. Deployed mixed-precision INT8 qZACH-ViT with Adam improves the FP32 ZACH-ViT baseline mean on all seven datasets, with a mean paired gain of 0.0313 in the dataset-specific primary metric; qZACH-ViT with RASO yields a mean gain of 0.0368. Across 964,920 source-to-INT8 test comparisons, prediction agreement is 99.9751%, with a mean absolute primary-metric change of 0.000133 and a maximum of 0.004386. Across 3,600 matched intrinsic maps, mean cosine similarity is 0.999955, mean rank correlation is 0.9944, and mean top-10% overlap is 0.9692. ONNX artifacts are 70.0% smaller than source checkpoints and provide 1.41× and 2.39× end-to-end CPU speedups with one and four threads. RASO significantly reduces sufficiency error and improves input-noise stability over Adam with the same attribution loss, but does not dominate every predictive or explainable artificial intelligence (XAI) metric. These results establish qZACH-ViT as a deployable compact intrinsically explainable model and RASO as a targeted stability-oriented optimization procedure.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 39 of its 48 references that are in the map. First posted on arXiv 16 Jul 2026.

Builds on

39 of 48 references.

References · 39

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  2. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  3. 2020Training data-efficient image transformers & distillation through attentioncited by 4,586
  4. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  5. 2017Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inferencecited by 1,593
  6. 2020Gradient Surgery for Multi-Task Learningcited by 1,256
  7. 2018Sanity Checks for Saliency Mapscited by 1,156
  8. 2020Concept Bottleneck Modelscited by 1,032
  9. 2019Attention is not Explanationcited by 929
  10. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 838
  11. 2017GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networkscited by 834
  12. 2018Multi-Task Learning as Multi-Objective Optimizationcited by 821
27 more
  1. 2021A Survey of Quantization Methods for Efficient Neural Network Inferencecited by 815
  2. 2020Quantifying Attention Flow in Transformerscited by 775
  3. 2018This Looks Like That: Deep Learning for Interpretable Image Recognitioncited by 737
  4. 2021MedMNIST v2 -- A large-scale lightweight benchmark for 2D and 3D biomedical image classificationcited by 690
  5. 2020Transformer Interpretability Beyond Attention Visualizationcited by 627
  6. 2018Towards Robust Interpretability with Self-Explaining Neural Networkscited by 505
  7. 2018A Benchmark for Interpretability Methods in Deep Neural Networkscited by 476
  8. 2021Conflict-Averse Gradient Descent for Multi-task Learningcited by 459
  9. 2017Interpretation of Neural Networks is Fragilecited by 444
  10. 2017The (Un)reliability of saliency methodscited by 333
  11. 2019On the (In)fidelity and Sensitivity for Explanationscited by 310
  12. 2019Explanations can be manipulated and geometry is to blamecited by 216
  13. 2021PTQ4ViT: Post-training quantization for vision transformers with twin uniform quantizationcited by 179
  14. 2022Quantus: An Explainable AI Toolkit for Responsible Evaluation of Neural Network Explanations and Beyondcited by 171
  15. 2020Gradient Vaccine: Investigating and Improving Multi-task Optimization in Massively Multilingual Modelscited by 169
  16. 2021FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformercited by 151
  17. 2022RepQ-ViT: Scale Reparameterization for Post-Training Quantization of Vision Transformerscited by 137
  18. 2022Q-ViT: Accurate and Fully Quantized Low-bit Vision Transformercited by 108
  19. 2022I-ViT: Integer-only Quantization for Efficient Vision Transformer Inferencecited by 95
  20. 2022Metrics for saliency map evaluation of deep learning explanation methodscited by 40
  21. 2022Q-ViT: Fully Differentiable Quantization for Vision Transformercited by 37
  22. 2023B-cos Alignment for Inherently Interpretable CNNs and Vision Transformerscited by 31
  23. 2024On the Faithfulness of Vision Transformer Explanationscited by 19
  24. 2023Holistically Explainable Vision Transformerscited by 8
  25. 2026ZACH-ViT: Regime-Dependent Inductive Bias in Compact Vision Transformers for Medical Imagingcited by 7
  26. 2026Extending ZACH-ViT to Robust Medical Imaging: Corruption and Adversarial Stress Testing in Low-Data Regimescited by 4
  27. 2025ZACH-ViT: A Zero-Token Vision Transformer with ShuffleStrides Data Augmentation for Robust Lung Ultrasound Classificationcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Quantifying Explanation Consistency: The C-Score Metric for CAM-Based Explainability in Medical Image Classification
  2. 2025MedVLM-R1: Incentivizing Medical Reasoning Capability of Vision-Language Models (VLMs) via Reinforcement Learning
  3. 2025Towards Efficient Medical Reasoning with Minimal Fine-Tuning Data
  4. 2026VQ-Seg: Vector-Quantized Token Perturbation for Semi-Supervised Medical Image Segmentation
  5. 2026Similarity-as-Evidence: Calibrating Overconfident VLMs for Interpretable and Label-Efficient Medical Active Learning
  6. 2025QoQ-Med: Building Multimodal Clinical Foundation Models with Domain-Aware GRPO Training
  7. 2025Resource-efficient medical image classification for edge devices
  8. 2024Efficient Quantization-Aware Training on Segment Anything Model in Medical Images and Its Deployment