{"total":11,"items":[{"citing_arxiv_id":"2606.24172","ref_index":43,"ref_count":1,"confidence":0.5,"is_internal_anchor":false,"paper_title":"A P\\={a}ninian Foundation for Indic Language Processing","primary_cat":"cs.CL","submitted_at":"2026-06-23T05:53:37+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Proposes treating Pāṇini's Astādhyāyī as a unifying computational architecture and benchmark foundation for Indic language NLP to improve accuracy, data efficiency, and transfer.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.18532","ref_index":32,"ref_count":1,"confidence":0.5,"is_internal_anchor":false,"paper_title":"AI Sandboxes: A Threat Model, Taxonomy, and Measurement Framework","primary_cat":"cs.CR","submitted_at":"2026-06-16T22:57:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"The paper presents a threat model, taxonomy, and six-dimension measurement framework for AI sandboxes to clarify valid testing claims for safety, security, and regulatory assurance.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.14668","ref_index":16,"ref_count":1,"confidence":0.5,"is_internal_anchor":false,"paper_title":"When to Write and When to Suppress: Route-Specialized Dual Adapters for Memory-Assisted Knowledge Editing","primary_cat":"cs.LG","submitted_at":"2026-06-12T17:37:32+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.09781","ref_index":55,"ref_count":1,"confidence":0.5,"is_internal_anchor":false,"paper_title":"Parameter-Efficient Neuroevolution for Diverse LLM Generation: Quality-Diversity Optimization via Prompt Embedding Evolution","primary_cat":"cs.NE","submitted_at":"2026-05-10T22:00:15+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Evolution strategies have demonstrated scalability to high- dimensional neural parameter spaces. Natural Evolution Strate- gies (NES) [63] provides theoretical grounding for natural gradient- based evolutionary optimization. OpenAI's work [52] showed ES can scale to millions of parameters, matching reinforcement learn- ing performance on complex tasks. Deep neuroevolution [55] demonstrated that genetic algorithms can evolve networks with over 4 million parameters, establishing the viability of gradient- free optimization at scale. Our approach shares this philosophy: prompt embeddings serve as a compact (∼32K parameters) indirect encoding influencing behavior of a much larger network (70B+ parameters). Rather than evolving the full network, we evolve the conditioning signal."},{"citing_arxiv_id":"2605.08302","ref_index":23,"ref_count":1,"confidence":0.5,"is_internal_anchor":false,"paper_title":"SGC-RML: A reliable and interpretable longitudinal assessment for PD in real-world DNS","primary_cat":"cs.LG","submitted_at":"2026-05-08T12:10:08+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SGC-RML creates an 8D symptom atlas from multimodal PD data and integrates conformal calibration to deliver reliable, rejectable longitudinal assessments.","context_count":1,"top_context_role":"method","top_context_polarity":"use_method","context_text":"064 ), so the apparent ACC drop isnotstatistically significant; the maintenance (or marginal improvement) of AUC confirms that the model retains stable ranking and discrimination capability on unseen subjects. Table 5.PADS visible 5-fold CV vs. hidden 50-subject blind test. ECE: 15-bin post temperature-scaling; Brier: 1 N P i,c(ˆpi,c −y i,c)2; ICC(2,1) [23]; Cov.: split-conformal coverage at target 0.80. Square brackets give Wilson 95% CIs for ACC and F1 (Hanley-McNeil for AUC); CIs of visible and hidden cohorts overlap, so the apparent ACC drop (−0.064) is not statistically significant at the 95% level. SettingnACC F1AUCECE↓Brier↓ICCCov.0.80 Visible_5fold_CV 419 0.704 [.659,.746] 0.588 [.541,.633] 0."},{"citing_arxiv_id":"2604.27263","ref_index":11,"ref_count":2,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Decoupling the Benefits of Subword Tokenization for Language Model Training via Byte-level Simulation","primary_cat":"cs.CL","submitted_at":"2026-04-29T23:29:08+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Byte-level simulations show subword tokenization improves LLM training mainly via increased throughput and boundary priors.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Tokenization is an essential step of the Natural Language Processing pipeline, segmenting text into atomic units to be processed by language models. Although state-of-the-art Large Language Models (LLMs) rely almost exclusively on subword algorithms like BPE or Unigram [31, 18], there is no consensus on which specific properties of subword models enable this performance advantage [11, 30]. Subword tokenization simultaneously dictates the allocation of compute to parts of the input se- quence and the scaling of the model's vocabulary parameters by balancing vocabulary size, se- quence length, and information density per token through the granularity of the tokens, orfertility of the tokenizer. Empirical evidence suggests that a larger vocabulary results on average in better"},{"citing_arxiv_id":"2604.25618","ref_index":15,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Beyond Isolated Utterances: Cue-Guided Interaction for Context-Dependent Conversational Multimodal Understanding","primary_cat":"cs.MM","submitted_at":"2026-04-28T13:24:38+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"CUCI-Net abstracts context-utterance dependency into an interpretation cue that combines local modality signals with global context and feeds it into the final multimodal interaction for context-conditioned predictions.","context_count":1,"top_context_role":"dataset","top_context_polarity":"use_dataset","context_text":"tARD to further test its generalization under binary humor and Preprint, 2026, arXiv Pan, Zhou et al. sarcasm recognition settings. Following prior work, the model pre- dicts the humor label on UR-FUNNY and the sarcasm label on MUStARD. Acc-2 is reported for both datasets under their standard evaluation protocols. Datasets.UR-FUNNY is a multimodal humor detection dataset proposed by Hasanet al.[ 15], containing 16,514 utterance instances from 1,866 TED talks with textual, acoustic, and visual modalities. MUStARD is a multimodal conversational sarcasm detection dataset proposed by Castroet al.[ 5], containing 690 dialogue instances with contextual segments and target utterances annotated with sarcasm labels. Compared with utterance-level sentiment analysis,"},{"citing_arxiv_id":"2604.21305","ref_index":25,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"WPGRec: Wavelet Packet Guided Graph Enhanced Sequential Recommendation","primary_cat":"cs.IR","submitted_at":"2026-04-23T05:44:26+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"WPGRec is a new sequential recommender that performs multi-scale temporal modeling via stationary wavelet packets and injects high-order collaborative information through scale-aligned graph propagation with energy-aware gated fusion.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.18988","ref_index":16,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"A Multi-Agent Framework with Structured Reasoning and Reflective Refinement for Multimodal Empathetic Response Generation","primary_cat":"cs.CV","submitted_at":"2026-04-21T02:18:18+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"A multi-agent framework decomposes multimodal empathetic response generation into structured reasoning steps and uses global reflection to reduce emotional biases, outperforming prior methods on IEMOCAP and MELD benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.17677","ref_index":4,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Semantic Entanglement in Vector-Based Retrieval: A Formal Framework and Context-Conditioned Disentanglement Pipeline for Agentic RAG Systems","primary_cat":"cs.AI","submitted_at":"2026-04-20T00:24:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"The paper introduces a measure of semantic entanglement in embeddings and a pipeline that improves Top-K retrieval precision from 32% to 82% on a healthcare knowledge base.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.20206","ref_index":85,"ref_count":1,"confidence":0.5,"is_internal_anchor":false,"paper_title":"PrivacyAkinator: Articulating Key Privacy Design Decisions by Answering LLM-Generated Multiple-choice Questions","primary_cat":"cs.HC","submitted_at":"2026-04-08T00:49:29+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"PrivacyAkinator uses LLM-generated questions grounded in data-flow representations and a news-mined design space to help developers surface privacy decisions, yielding 47% more decisions identified in 73% less time than PRAM in a 24-person study.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Privacy Design Space. The complexity of privacy design has driven researchers to develop structured approaches for mapping out the privacy design space, particularly for notice and choice mechanisms [30, 85]. For example, Schaub et al. proposed a design space for privacy notices by identifying key dimensions such as tim- ing, modality, and channel [85]. While these efforts provide usable taxonomies and vocabularies for categorizing and communicating different privacy designs, they focus narrowly on notice and user control rather than the broader privacy design landscape. In parallel, there has been extensive work on the analysis of privacy policies [22, 38, 87, 94, 115]. Wilson et al. developed the"}],"limit":50,"offset":0}