{"total":14,"items":[{"citing_arxiv_id":"2607.01648","ref_index":24,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Boosting Ultrasound Image Classification via Attribute-Guided Dual-Branch Framework","primary_cat":"cs.CV","submitted_at":"2026-07-02T03:20:12+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":5.0,"formal_verification":"none","one_line_summary":"An attribute-guided dual-branch framework fuses a standard classifier with an interpretable attribute-prior branch to boost ultrasound classification accuracy and explainability.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.30901","ref_index":28,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"GRAPE: Graph-Augmented Prototype Explanations for Interactive Medical Image Diagnosis","primary_cat":"cs.CV","submitted_at":"2026-06-29T20:45:26+00:00","verdict":"UNVERDICTED","verdict_confidence":"UNKNOWN","novelty_score":6.0,"formal_verification":"none","one_line_summary":"GRAPE augments prototype medical image classifiers with graph attention for co-occurrence, a mismatch safety check, and open-vocabulary anchoring to support incremental addition of findings from single examples.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.19741","ref_index":19,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Interpreting Neural Combinatorial Optimization via Evolving Programmatic Bottlenecks","primary_cat":"cs.AI","submitted_at":"2026-06-18T03:14:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"EPB distills NCO models into evolving program portfolios via LLM-driven textual-numerical optimization, matching original performance while exposing stage-dependent heuristic-like behavior.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.19489","ref_index":20,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Concept Flow Models: Anchoring Concept-Based Reasoning with Hierarchical Bottlenecks","primary_cat":"cs.LG","submitted_at":"2026-06-17T18:27:17+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Concept Flow Models use hierarchical concept-driven decision trees to mitigate information leakage in concept bottleneck models while matching their predictive performance.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.11446","ref_index":3,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"3D-CBM: A Framework for Concept-Based Interpretability in Generative 3D Modeling","primary_cat":"cs.CV","submitted_at":"2026-06-09T20:57:03+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Introduces 3D-CBM framework mapping raw 3D inputs to multi-tiered interpretable concepts, achieving 88.8% concept accuracy and test-time intervention on PartNet and ShapeNet.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.04326","ref_index":99,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models","primary_cat":"cs.LG","submitted_at":"2026-06-03T01:01:05+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Introduces synthetic benchmarks for concept bottleneck models that control data modality, concept choice, annotation quality, and completeness to evaluate performance in decision support and automation.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.01698","ref_index":20,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Learning Label-Efficient Interpretable Medical Image Diagnosis via Semi-supervised Hypergraph Concept Bottleneck Model","primary_cat":"cs.CV","submitted_at":"2026-06-01T05:05:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A new semi-supervised hypergraph Concept Bottleneck Model framework improves label efficiency and interpretability for medical image diagnosis on PAS ultrasound, breast ultrasound, and SkinCon datasets.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.25304","ref_index":29,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"When Interpretability Becomes a Liability: Adversarial Attacks on CBM Concept Layers","primary_cat":"cs.LG","submitted_at":"2026-05-25T00:03:47+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Concept-level adversarial attacks exploit CBM interpretability on the CUB dataset, but SPECTRA raises required perturbation norm from 0.46 to over 4200 while keeping accuracy loss under 2.2%.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15688","ref_index":82,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"$\\alpha$-TCAV: A Unified Framework for Testing with Concept Activation Vectors","primary_cat":"stat.ML","submitted_at":"2026-05-15T07:22:59+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"α-TCAV replaces TCAV's hard indicator with a tunable smooth function to create a unified probabilistic framework with lower variance and guidance for parameter choice or Bayes-optimal scoring.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12639","ref_index":45,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"OceanCBM: A Concept Bottleneck Model for Mechanistic Interpretability in Ocean Forecasting","primary_cat":"cs.LG","submitted_at":"2026-05-12T18:29:45+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"OceanCBM is the first concept bottleneck model for spatiotemporal ocean prediction that uses mixed supervision on physical concepts and a free concept to deliver consistent mechanistic representations for mixed layer heat content forecasts.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.11558","ref_index":81,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"A Composite Activation Function for Learning Stable Binary Representations","primary_cat":"cs.LG","submitted_at":"2026-05-12T05:41:36+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"HTAF is a sigmoid-tanh composite that approximates the Heaviside function to allow stable gradient training of binary activation networks, yielding ICBMs with stable discretization and competitive performance on image tasks.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"[79] Chang Yue and Niraj K Jha. Learning interpretable differentiable logic networks.IEEE Transactions on Circuits and Systems for Artificial Intelligence, 2024. [80] Mert Yuksekgonul, Federico Bianchi, Pratyusha Kalluri, Dan Jurafsky, and James Zou. When and why vision-language models behave like bags-of-words, and what to do about it?arXiv preprint arXiv:2210.01936, 2022. [81] Mert Yuksekgonul, Maggie Wang, and James Zou. Post-hoc concept bottleneck models.arXiv preprint arXiv:2205.15480, 2022. [82] Sergey Zagoruyko and Nikos Komodakis. Wide residual networks.arXiv preprint arXiv:1605.07146, 2016. [83] Friedemann Zenke and Surya Ganguli. Superspike: Supervised learning in multilayer spiking neural networks.Neural computation, 30(6):1514-1541, 2018."},{"citing_arxiv_id":"2604.19323","ref_index":5,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Concept Inconsistency in Dermoscopic Concept Bottleneck Models: A Rough-Set Analysis of the Derm7pt Dataset","primary_cat":"cs.LG","submitted_at":"2026-04-21T10:45:50+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Rough-set analysis finds 16.4% of 305 concept profiles in Derm7pt inconsistent (306 images), capping hard CBM accuracy at 92.1%; symmetric filtering produces a 705-image consistent benchmark where EfficientNet-B5 reaches 0.90 label accuracy.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"signals that the dataset is inconsistent, meaning that no concept-based classifier can perfectly separate the two classes regardless of architecture, training procedure, or supervision scheme. Let Π denote the set of inconsistent concept profiles. For profile ek ∈Π , let nmel k and nnmel k be the melanoma and non- melanoma image counts, respectively, with total sizen k =n mel k +n nmel k . The conflict ratio is defined as follows: γk = min nmel k ,n nmel k \u0001 nk ,γ k ∈(0,0.5].(5) A value γk =0.5 indicates a perfectly ambiguous profile, while γk →0 indicates a near-consistent profile where a single annotation outlier creates the conflict. Let nmaj k =max(n mel k ,n nmel k ) be the majority-class count in profile k. Therefore, the concept-based accuracy ceiling is given by the following expression: acc∗ = |POSC(d)|+ ∑ k∈Π nmaj"},{"citing_arxiv_id":"2604.17089","ref_index":2,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Tree of Concepts: Interpretable Continual Learners in Non-Stationary Clinical Domains","primary_cat":"cs.LG","submitted_at":"2026-04-18T17:58:37+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Tree of Concepts uses a fixed rule-based concept interface from a shallow decision tree to support continual adaptation in clinical data while preserving consistent explanations across updates.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.02468","ref_index":9,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Hierarchical, Interpretable, Label-Free Concept Bottleneck Model","primary_cat":"cs.CV","submitted_at":"2026-04-02T19:02:59+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"HIL-CBM is a hierarchical label-free concept bottleneck model that improves classification accuracy and explanation quality over prior single-level CBMs using a visual consistency loss and dual heads.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"can understand why the model predicts a specific bird species based on the presence of visual characteristics associated with that species. Despite their interpretability, early CBMs face several limi- tations. They must be trained from scratch, which is computa- tionally expensive, and depend on the availability of annotated concept labels. To mitigate the retraining burden, Post-hoc CBM [9] is trained only on the final classifier and an optional residual fitting layer, leveraging Concept Activation Vectors (CA Vs) [10] and CLIP [11] to extract concept directions from pre-trained image encoders. However, this approach assumes access to the CLIP encoder and is mostly applicable to datasets without concept annotations. To address the annotation limitation, the Label-Free Concept"}],"limit":50,"offset":0}