Retrieval pipelines, vector search, chunking and reranking: how models are grounded in a corpus instead of their weights.
arXiv:2605.08802v4 Announce Type: replace Abstract: Due to the potential for exploratory reasoning of Latent Visual Reasoning, recent works tend to enable MLLMs (Multimodal Large Language Models) to...
arXiv:2610.29098v1 Announce Type: new Abstract: One-shot landmark localization in volume scans with varying fields of view (FOVs) is challenging. While existing self-supervised learning methods have...
arXiv:2610.29136v1 Announce Type: new Abstract: Continuous diffusion language models generate a whole sequence in parallel, never drawing a single word at any intermediate step; only the final step t...
The paper introduces Latent Space Configuration (LSC) for fine‑tuning neural networks, enabling the creation of structured latent spaces that accelerate top‑K retrieval with O(1) candidate search. Using ArcFace‑pretrained IResNet50 models, the authors demonstrate up to three‑fold GPU speed‑ups and 40% CPU speed‑ups over exhaustive search, while also showing that reducing latent space dimensionality further speeds up queries and can improve rank‑N metrics.
arXiv:2610.29722v1 Announce Type: new Abstract: Retrieval Augmented Generation (RAG) grounds large language models in external knowledge. Modern pipelines combine embedding, chunking, retrieval, rera...
Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection introduces OracleAD, an unsupervised framework that encodes each variable’s past into a causal embedding to jointly predict the present and reconstruct the input window. The embeddings are projected via self‑attention into a shared latent space, capturing dynamic spatial relationships that are aligned to a Stable Latent Structure (SLS) representing normal-state relationships. Anomalies are detected through a dual scoring mechanism based on prediction error and deviation from the SLS, allowing fine‑grained diagnosis and root‑cause identification at the variable level.
This paper addresses the problem of creating immersive volumetric videos from sparse and low-resolution capture resources. Conventional dense multi-view reconstruction methods can produce high-quality...
Learning across medical imaging modalities requires making use of images without paired text and accommodating new domains while retaining existing representations. We propose a medical vision-languag...
Text-to-Image models reproduce societal stereotypes, and existing mitigation applies one demographic correction to the whole scene. In multi-person scene generation this creates a trade-off: global st...
Graph-level out-of-distribution (OOD) generalization is critical for deploying graph neural networks in real-world applications where distribution shifts are inevitable. Existing approaches typically...
User-facing LLM agents must identify which behavioral policies apply at each turn, but evaluating every policy with an LLM is costly. We introduce Signals and Queries (S&Q), a lightweight first-stage...
arXiv:2610.27816v1 Announce Type: cross Abstract: Retrieval-Augmented Generation (RAG) has become the dominant paradigm for grounding the outputs of large language models (LLM) in external knowledge....
arXiv:2610.27555v1 Announce Type: cross Abstract: Safety pretraining has emerged as a promising paradigm for embedding safety deeply into a model during pretraining, yielding robustness to subsequent...
The paper introduces EmbGA, an embedding‑space genetic algorithm designed to uncover hidden concepts in text‑to‑image models that are not explicitly mentioned in prompts. Unlike existing prompt optimizers, EmbGA operates directly in token embedding space, using crossover and mutation to generate valid token sequences that align with unseen visual concepts. Experiments on two benchmarks show that EmbGA outperforms six other optimizers in alignment and token grounding, especially with larger exploration budgets, while also being faster than language‑model‑guided search.
LocaLeJEPA extends Joint-Embedding Predictive Architectures (JEPA) by introducing a manifold-theoretic reformulation that aligns representation learning with non‑Euclidean geometric priors. It incorporates a locally Euclidean distribution‑matching regularizer, LocaLeReg, to stabilize training and scale to large models. Experiments on ImageNet‑1K show that LocaLeJEPA outperforms the original LeJEPA across both hyperbolic and Euclidean backbones, improving linear and k‑NN classification as well as downstream tasks.
The paper investigates how dimensionality reduction and quantization together affect text embedding compression across four task families in the Massive Text Embedding Benchmark using five pretrained models. It finds that combining these techniques can preserve near-original performance while drastically reducing storage, with some classification models retaining at least 99% of their score using less than 0.1% of the original storage. The optimal trade‑off between dimensionality and bit‑width varies by task, with retrieval tasks being more sensitive to compression.
The paper introduces FADNet, a face authenticity detector that treats forgery detection as a one‑class classification problem, training only on authentic faces. It uses Evidential Deep Learning to measure uncertainty and a pseudo‑forgery generator to sharpen decision boundaries. Experiments on DF40 and ASFD show FADNet surpasses state‑of‑the‑art methods with 96.63% accuracy and 98.83% precision.
DriveVLA-M0 introduces a failure-aware latent memory system for Vision‑Language‑Action models in autonomous driving. It stores past failure cases with scene representations and expert trajectories, and uses a retrieval model to decouple static road structure from dynamic agent interactions. During inference, retrieved cases are injected via a lightweight LoRA‑based test‑time training mechanism, yielding significant performance gains on NAVSIM benchmarks with minimal latency overhead.
CondenseVLA introduces a learnable history‑condensation module that distills short‑term visual context into a fixed set of slot tokens via learnable queries and cross‑attention, reducing redundancy and improving action decoding. The approach achieves notable gains on several robot manipulation benchmarks, raising Simpler‑WidowX success from 58.7% to 69.8%, reaching 97.2% on LIBERO, 52.8% on real‑world tasks, and 21.6% on memory‑dependent RMBench tasks.
arXiv:2610.27941v1 Announce Type: new Abstract: AI-assisted dementia detection usually requires multimodal fusion of diverse data, such as clinical tabular assessments and neuroimaging data. However,...