Retrieval pipelines, vector search, chunking and reranking: how models are grounded in a corpus instead of their weights.
arXiv:2610.28337v1 Announce Type: cross Abstract: Emoji are widely treated as compact affective tokens whose meaning is summarised by a single sentiment score. We argue that this picture is incomplet...
arXiv:2610.28364v1 Announce Type: cross Abstract: Local Retrieval-Augmented Generation (RAG) systems are vulnerable to indirect prompt injection and retrieval poisoning, where malicious instructions...
arXiv:2610.28624v1 Announce Type: cross Abstract: Warehouse robot fleets continuously observe operational conditions, but preserving these observations as reusable semantic memory incurs substantial...
arXiv:2511.18354v2 Announce Type: replace-cross Abstract: AI agents, not humans, are becoming the web's primary consumers, yet the web still ships content structured for human browsing, with no nativ...
arXiv:2603.26246v2 Announce Type: replace-cross Abstract: Standard LLM-based speech recognition systems process utterances in isolation, limiting their ability to leverage conversational context. Her...
Long-horizon robotic operation requires memory that scales with essential information rather than execution time. We present a parameter-space memory framework where each episode is stored as a separa...
Learning latent dynamics is fundamental to building action-conditioned world models for planning. Among existing families, Joint Embedding Predictive Architectures (JEPAs) offer a natural framework by...
Endoscopic video contains state transitions induced by camera displacement, deformable tissue, and instrument interaction, yet existing pre-training methods treat dynamics implicitly. Inspired by acti...
The paper introduces a unified framework for measuring divergences between conditional distributions using kernel embeddings, termed conditional maximum mean discrepancy (CMMD). It defines a family of CMMD metrics—CMMD_0, CMMD_1, and CMMD_2—each based on different RKHS embeddings, and generalizes them to a level‑s CMMD with clear assumptions and operator‑based smoothing connections. Additionally, the authors present a novel doubly robust estimator for CMMD and demonstrate through experiments that the metric captures complex conditional dependencies for statistical testing.
IMPLICIT-Bench is a benchmark of 1,831 prompt triplets that tests whether text‑to‑image models can override stereotypical defaults when given explicit requests. Each triplet contains a neutral prompt, a stereotype prompt, and an anti‑stereotype prompt, allowing measurement of a model’s default bias and its ability to follow contradictory instructions. Experiments on Qwen‑Image and Stable Diffusion 3 show that models often default to stereotypes, succeed less often on anti‑stereotype requests, and that prompt rewriting or embedding steering can modestly reduce bias when the relevant stereotype is known.
The paper introduces the Recursive Time-Series Model (RTSM), a deep-learning architecture that uses a learned convolutional temporal tokenizer, a single-state weight‑tied refinement module, and a pooled classification readout to classify multivariate time‑series data. On the 30 UEA datasets, RTSM achieves a mean accuracy of 72.81% under validation‑based configuration, surpassing all 17 evaluated deep‑learning models and outperforming MambaSL by 1.60 percentage points. Ablation studies show that the proposed tokenizer and single‑state design outperform simpler convolutional embeddings and two‑state variants, and the full classifier reaches 86.10% mean accuracy compared to 84.79% for a tokenizer‑only classifier.
The paper introduces latent‑space language steering, a lightweight inference‑time technique that uses Principal Component Analysis on parallel translations to identify language directions in token embeddings. By steering embeddings along these axes, the method reduces unintended code‑switching in multilingual large language models, achieving up to 55% lower next‑token distributional divergence and a 63–99% drop in the Code‑Switching Index while maintaining semantic integrity and incurring negligible computational cost. The authors also demonstrate that a single principal component can classify language with 95–99% accuracy and provide an analysis of how language representations evolve across model layers.
FlexPainter is a new texture generation pipeline that offers flexible multi‑modal conditional guidance and produces highly consistent textures across multiple views. It builds a shared conditional embedding space to aggregate different input modalities and introduces an image‑based CFG method to separate structural and style information for reference image‑based stylization. The system generates multi‑view images simultaneously using a grid representation, applies view synchronization and adaptive weighting during diffusion sampling for local consistency, and finally uses a 3D‑aware texture completion model with a texture enhancement model to produce seamless, high‑resolution texture maps, outperforming state‑of‑the‑art methods in flexibility and quality.
SAG2 is an embedding‑free multi‑hop retrieval system that relies on a single SQLite database, BM25, and contextual evidence decisions from JEV. It replaces dense indexes with a lexical index over original passages, allowing query‑time evidence selection without offline embedding computation or dedicated database servers. On several benchmark datasets, SAG2 achieves the highest Recall@5 and Recall@10, demonstrating strong retrieval accuracy with lightweight preparation and rapid indexing.
FaithfulFaces introduces a pose‑faithful facial identity preservation framework for text‑to‑video generation, addressing identity distortion caused by large pose variations or occlusions. The method employs a pose‑shared identity aligner that refines and aligns facial poses across views using a pose‑shared dictionary and a pose‑variation‑identity invariance constraint. By mapping single‑view inputs into a global pose representation with explicit Euler angle embeddings, the framework provides a robust facial prior that guides generation toward consistent identity and structural clarity, achieving state‑of‑the‑art performance on a curated high‑quality video dataset.
The paper investigates position bias in large language models (LLMs) across a wide range of model sizes—from billions to trillions of parameters—and across five languages. It finds that while small-scale models and non‑English languages still exhibit noticeable position bias, larger models and the use of Chain‑of‑Thought prompting substantially reduce this bias. Additionally, the study shows that emerging thinking models with advanced reasoning capabilities are almost immune to position bias, indicating a promising but resource‑intensive path to mitigating this vulnerability.
The paper introduces a framework that aligns multimodal embeddings with language by using captioning functions on interpretable auxiliary data, a soft contrastive loss, and a rank-based concept retrieval task. Applied to Earth observation embeddings, the method improves species distribution modelling, achieving state‑of‑the‑art performance on most metrics and better concept retrieval than prior vision‑language models. The aligned concepts enable recovery of species‑habitat associations and allow domain experts to interpret model predictions in natural language.
Atomic Consistency Preference Optimization (ACPO) is a self‑supervised preference‑tuning method that improves factual accuracy in large language models without external supervision. It uses atomic consistency signals—agreement of individual facts across multiple stochastic responses—to identify high‑ and low‑quality data pairs for model alignment. ACPO outperforms a strong supervised alignment baseline by 1.95 points on the LongFact and BioGen datasets for Phi‑3 and Llama‑3 models.
Optimist Deep Research is an open‑source agent that conducts literature research as an iterative, state‑centric process. It breaks a research question into a tree of subquestions, uses search intents to query multiple scholarly databases, and merges results into a provenance‑preserving document registry. The agent iteratively expands the registry via citation snowballing, evaluates evidence, and refines the research plan through a critic that can add, split, rephrase, deprioritize, or close subquestions, ultimately producing a report with every citation linked to a retrieved document.
SPARE is a framework for streaming video understanding that adaptively retrieves historical evidence when recent observations are insufficient to answer a question. It uses a lightweight gate to decide when to fetch relevant past clips via a multimodal embedding model, and trains the system with LoRA instruction tuning to jointly use recent and retrieved context while abstaining when evidence is lacking. By sharing a visual encoder between understanding and retrieval branches and aligning features with linear adapters, SPARE achieves state‑of‑the‑art accuracy on OVO‑Bench and StreamingBench, outperforming Qwen3‑VL‑8B‑Instruct by over 20 percentage points on OVO‑Bench and 14 on StreamingBench.