Retrieval pipelines, vector search, chunking and reranking: how models are grounded in a corpus instead of their weights.
Using a Large Language Model (LLM) as the clusterer at production scale is hard: prompts cannot hold the entire label space, and per-document serial processing does not deliver the throughput real wor...
Brain-to-image retrieval seeks to identify the visual stimulus that elicited a non-invasive neural response. Candidate images are typically represented by pretrained vision models, whose internal repr...
The paper introduces the Memory Decision Layer (MDL), a zero‑parameter controller that sits between retrieval and generation in large language models. MDL uses a three‑signal complementary encoder—combining relevance, reliability, and task risk—to produce an interpretable decision about the trustworthiness of retrieved memories. By decoupling confidence from consistency and enabling risk inversion and abstention, MDL cuts hallucination rates under conflicting memories by roughly 56% and nearly eliminates them in high‑risk scenarios, all while adding only 0.14 ms per decision.
PRIME introduces a perception feedback mechanism for Vision‑Language‑Action models in autonomous driving, conditioning perceptual queries on a Situational Memory that aggregates past perception, reasoning, navigation goals, and predicted behaviors via cross‑attention. This approach adds only 29.7 M parameters (0.41 % of a 7.3 B‑parameter base model) and enables intent‑driven perceptual attention at minimal computational cost. On the Bench2Drive closed‑loop benchmark, PRIME achieves a state‑of‑the‑art Driving Score of 82.47 and a Success Rate of 60.00 %, outperforming prior models such as ORION.
The paper introduces an end‑to‑end framework for personalized audio‑driven facial motion that operates in real time without look‑ahead. It combines a causal multi‑resolution motion tokenizer, which captures both global temporal context and fine articulatory details, with a multi‑modal style retriever that pulls stylistic priors from arbitrary reference footage using ongoing audio and motion queries. This approach allows high‑fidelity, identity‑consistent animation from just a few casually recorded clips, outperforming existing methods in lip‑sync accuracy, identity consistency, and perceived realism while maintaining real‑time streaming constraints.
COMPLEX is a closed‑form, training‑free embedding for multiparameter persistence modules that provides both an upper and a lower Lipschitz bound, enabling faithful feature representations. By slicing modules along a near‑diagonal net and embedding each slice with the certified PLACE/PALACE landmark map, the method guarantees that separated modules remain separated in the embedding. On Orbit benchmarks and molecular graph tasks, COMPLEX achieves state‑of‑the‑art accuracy, outperforming existing landmark, transformer, and graph‑based approaches.
The paper argues that zero‑shot time‑series forecasting should be treated as an evidence‑access claim rather than merely a no‑parameter‑update condition. It introduces a source‑first taxonomy that distinguishes three evidence sources—frozen LLM prior reuse, parametric time‑series pretraining, and retrieval‑augmented external memory—from the architectures that implement them. The authors further outline four audit questions—task interface, forecast object and scoring, prediction‑time context, and resource budget—to make zero‑shot leaderboards transparent and comparable.
MemeTAG introduces a dual‑objective framework for classifying harmful memes by combining keyword generation from a pretrained Vision‑Language Model with an Aggregated Tag Inference Network (ATIN) that condenses these keywords into a rich semantic embedding. The embedding is used as a target for an auxiliary reconstruction loss, encouraging deep alignment between visual and textual features. This approach, along with a three‑stage training strategy, achieves new state‑of‑the‑art results on the HarMeme, Hateful Memes Challenge, and PrideMM datasets.
The paper presents an AI-based methodology that uses embedding-based similarity search and large language model classification to map Dutch job profiles to harmonised ESCO occupations. It introduces a Digital Semantic Score, which quantifies the association of job titles and skills with digital concepts relative to a non-digital reference, moving beyond keyword-based approaches. The study finds that digitalisation is unevenly distributed across occupations, with managerial, professional, and ICT roles showing the strongest digital language, while career-transition and skill analyses reveal pathway-dependent movement toward digital work and a multidimensional nature of digital capability.
The paper introduces a framework to assess how linearly encoded linguistic relations are within language model embeddings. Using a constrained linear approximation on related and unrelated word pairs, the authors evaluate an extended BATS dataset across inflectional, derivational, lexicographic, and encyclopedic relations in GloVe, RoBERTa, and ModernBERT. Results show near-perfect linear encoding for inflectional and derivational relations, higher errors for lexicographic and encyclopedic relations—especially in one-to-many and many-to-many cases—and that RoBERTa and ModernBERT encode relations more linearly than GloVe.
The paper introduces SceneLM, a vision‑language model that maintains an open‑vocabulary 3D scene map using only a structured text list of objects as persistent memory. The model updates this textual map by adding, editing, and removing objects for each input image, learning the process through supervision tasks and an automatic annotation pipeline. Evaluations on language‑grounded retrieval and localization benchmarks show competitive performance with traditional mapping systems while producing a 6‑12× more compact representation, and the model can run online on an edge device such as a quadruped robot.
The paper evaluates pre‑trained speech embeddings from four state‑of‑the‑art speech foundation models for cross‑lingual Parkinson's disease severity assessment. Experiments span three datasets in zero‑shot and k‑shot settings, showing that these embeddings can transfer meaningfully across languages, though performance varies with dataset characteristics, preprocessing, and adaptation strategy. Misclassifications linked to inter‑speaker variability and atypical speech patterns underscore the need for more robust feature extraction, modeling, and explainability to support reliable clinical insights.
The paper introduces techniques for efficient on-device diffusion-based image generation on consumer GPUs. It presents an embedding translator that reduces weight and latency by mapping a small text encoder into a larger encoder space, a reproducible sweep recipe for balancing speed, quality, and memory, and an interactive editor that achieves sub‑second time‑to‑first‑image on recent GPUs. These contributions aim to broaden the reach of diffusion pipelines to a wide range of client devices.
The paper introduces Prototype Purification and Regulation (PPR), a multi‑label few‑shot learning framework for medical image classification that addresses two key limitations of existing metric‑based meta‑learning methods. PPR first purifies prototypes by using sample‑level comorbidity scores to highlight disease‑specific features, then regulates inter‑class prototype distances with disease‑level comorbidity statistics to create a comorbidity‑aware embedding space. Experiments on four chest X‑ray datasets, including cross‑domain tests, show that PPR outperforms state‑of‑the‑art methods, improving disease detection and demonstrating robust generalization and clinical applicability.
The paper presents an iterative framework that uses large language models (LLMs) to automatically extract interpretable, schema‑bound categorical features from unstructured text for use in tabular prediction models. A generator LLM proposes semantic definitions, an extractor LLM materializes the features, and a downstream tabular model evaluates their predictive performance, with error‑driven natural‑language feedback guiding the search. Across three public datasets, the error‑driven loop speeds up feature discovery up to three times and the resulting features outperform any subset when combined with TF‑IDF and dense embeddings, while also providing instance‑level interpretability through SHAP importance rankings and a semantic audit trail.
arXiv:2510.24233v2 Announce Type: replace Abstract: Deep generative models are often trained on sensitive data, such as genetic sequences, health data, or more broadly, any copyrighted, licensed or p...
arXiv:2511.14301v4 Announce Type: replace-cross Abstract: Transformer-based models are highly susceptible to backdoor attacks via supervised fine-tuning (SFT). To red-team existing data-poisoning def...
arXiv:2609.21138v1 Announce Type: cross Abstract: Octopus crawling motivates soft robots that exploit redundancy, yet discovering and organizing diverse coordination modes for adaptation remains chal...
The paper introduces a two‑step test‑time training protocol, Embed‑TTT, for Vision ARC (VARC) that first fine‑tunes only the task embedding and then fine‑tunes the backbone. This approach consistently produces task embeddings that better align with the underlying rules, improves retrieval and linear probing, and recovers the geometric structure of parametric rules. Even fine‑tuning only the tiny embedding component solves a significant portion of ARC‑AGI‑1, ConceptARC, and Mini‑ARC tasks, while the full two‑step pipeline further enhances performance and demonstrates compositional rule interpolation.
The paper explores offline large language models as decision‑support tools for air operations, where analysts must work with doctrine and imagery under limited connectivity. It presents a modular retrieval‑augmented architecture that handles both text and image inputs from technical manuals and can operate without Internet access. A pilot study with four Brazilian Air Force image analysts shows the system matches human performance on doctrinal knowledge tests while completing the task in a fraction of the time, and it highlights the high cognitive load of manual target reporting.