๐ Information Retrieval & RAG¶
๐ง NeurIPS2026 ยท 18 paper notes
๐ Same area in other venues: ๐๏ธ ECCV2026 (8) ยท ๐ฌ ICLR2026 (81) ยท ๐ฌ ACL2026 (73) ยท ๐งช ICML2026 (26) ยท ๐ค AAAI2026 (21) ยท ๐ง NeurIPS2025 (25)
๐ฅ Top topics: RAG ร3
- Beyond Raw Context Transfer: Representation-based Federated Retrieval-Augmented Generation
-
Addressing the severe privacy risks and lengthy context overhead of exchanging raw documents in decentralized environments, FedRepRAG keeps raw text localized and exchanges only compact latent embeddings, leveraging a collaboratively trained lightweight projector to inject each retrieved document as a single soft token into frozen LLM/VLM backbones, slashing context length by 19โ398ร while substantially outperforming local retrieval.
- Can Language Models Actually Retrieve In-Context? Drowning in Documents at Million Token Scale
-
Focusing on corpus-scale in-context retrieval (ICR) at million-token context lengths, this paper diagnoses the root cause of retrieval collapse as attention dilution, and introduces length-aware Softmax rescaling (SSMax) alongside document-level sparse routing to match a same-backbone dense retriever.
- Cluster with Auctions for Vector Search
-
Decouples database partitioning from query probing in inverted indexes, optimizing cluster assignments via GPU-parallel capacitated auction algorithms and query probing via neural networks alternatively, yielding up to 4.7ร throughput over competitive baselines under severe distribution shift.
- Data Auctions for Retrieval Augmented Generation
-
Addressing the need for exclusive selling of sensitive knowledge to GenAI platforms for RAG tasks, this paper formulates a model-agnostic coverage valuation function, proves that welfare maximization and demand queries are NP-hard, and designs the LPRMono mechanismโcombining LP relaxation, randomized rounding, and a novel "data burning" post-processing stepโto achieve dominant-strategy truthfulness with a tight \((1-1/e)\) approximation ratio.
- Diagnosing and Repairing Citation Failures in Generative Engine Optimization
-
This paper reframes Generative Engine Optimization (GEO) from generic text rewriting into a diagnostic repair paradigm, proposing a full-pipeline taxonomy of citation failures alongside AgentGEO, an agentic framework that achieves over 40% relative citation rate improvement while modifying merely 5% of webpage content.
- From Topology to Retrieval: Decoding Embedding Spaces with Unified Signatures
-
This paper introduces Unified Topological Signatures (UTS), a multidimensional geometric framework aggregating persistent homology, intrinsic dimensionality, diversity, isotropy, and clustering descriptors to comprehensively characterize text embedding spaces, accurately predict downstream dense retrieval effectiveness and document retrievability bias without ground-truth labels, and refute scale-driven representation convergence claimed by the Platonic Representation Hypothesis.
- GISA: A Benchmark for General Information Seeking Assistant
-
Addressing key flaws in current search agent benchmarksโsuch as reverse-engineered queries, artificial separation of depth and breadth, static answers vulnerable to pre-training contamination, and unstable LLM judgesโGISA introduces a human-crafted benchmark featuring four structured answer types, dynamic live subsets, and complete human search trajectories for process-level supervision.
- Is Dimensionality a Barrier for Retrieval Models?
-
This paper theoretically investigates whether representation dimensionality limits large-scale retrieval, proving that near-optimal retrieval margins are achievable in dimensions scaling only logarithmically with dataset size, \(d = O(\mathsf{m}^{\mathrm{rd}}(+\infty, A)^{-2}\log(\min(n, N)))\), while revealing that Sigmoid loss significantly surpasses InfoNCE for large-margin low-dimensional geometry.
- Knowledge-Graph Paths as Intermediate Supervision for Self-Evolving Search Agents
-
Addressing the twin bottlenecks of ungrounded question synthesis in Proposers and sparse binary rewards in Solvers within Search Self-Play (SSP), this paper introduces construction-derived intermediate supervision by reusing knowledge-graph paths to both provide relational context for multi-hop question generation and supply non-zero partial credit to incorrect Solver trajectories via Waypoint Coverage Reward.
- LeanSearch v2: Global Premise Retrieval for Lean 4 Theorem Proving
-
Addressing the challenge of locating scattered, non-vocabulary-overlapping supporting lemmas in Lean 4 formalization, LeanSearch v2 introduces a two-mode system combining a hierarchy-informalized embeddingโreranker substrate (standard mode) and an iterative sketch-retrieve-reflect loop (reasoning mode), substantially advancing global premise recall and downstream proof success.
- Learning to Align Generative Appearance Priors for Fine-grained Image Retrieval
-
To overcome label-driven discriminative overfitting to seen-category semantics in fine-grained image retrieval, GAPan leverages an invertible normalizing flow to map visual features to class-conditional Gaussian priors and synthesizes appearance-aware anchors to guide contrastive alignment, substantially boosting generalization on unseen categories.
- LogicTree-RAG: Logic Tree-guided Retrieval-Augmented Generation for Long-form Patent Drafting
-
LogicTree-RAG organizes a research paper into a retrievable, recursively expandable logic tree with evidence associations, then uses section-specific hybrid traversal to generate a long patent-description draft, achieving 20.22k output tokens, 43.79 coverage, and 66.34 source-combined factuality on the Pap2Pat test set; these results do not certify patentability or legal validity.
- MaxSketch: Robust Distinct Counting in Streams via Random Projections
-
Addressing the challenge of robust distinct counting in high-dimensional data streams where repeated instances of the same object are only approximately similar, this paper introduces MaxSketch, a max-linear sketch built from random Gaussian projections that achieves \((1+\varepsilon)\) multiplicative approximation using \(\tilde{O}(\log n / \varepsilon^2)\) memory under natural clustered geometry, bypassing the \(\Omega(\sqrt{n})\) lower bound for general metric spaces.
- MCGI: Manifold-Consistent Graph Indexing for Billion-Scale Disk-Resident Vector Search
-
Addressing the search degradation in disk-resident graph indices caused by the divergence between Euclidean distance and manifold geodesics in high-dimensional spaces, this paper introduces Manifold-Consistent Graph Indexing (MCGI), which leverages Local Intrinsic Dimensionality (LID) to dynamically modulate edge pruning constraints, achieving a 5.8x throughput gain on 960-dimensional GIST1M and a 3x latency reduction on billion-scale SIFT1B.
- Pinpoint: Grounded Worldwide Image Geolocation via Cross-Source Retrieval and Reranking
-
Pinpoint introduces an MLLM-free retrieve-and-rerank framework for worldwide image geolocation that joint-trains contrastive embeddings across Flickr photos and street-view imagery, grounding top candidates with local support tokens to achieve state-of-the-art accuracy with 3000x faster inference.
- Semantic Search over 9 Million Mathematical Theorems
-
By extracting a 9.2-million theorem corpus from arXiv and 7 open-source texts, this work formulates theorem retrieval as an asymmetric task via LLM-generated natural-language slogans and a two-stage vector index, achieving a 45.0% theorem-level Hit@20 on expert mathematician queries.
- UniRank: Unified List-wise Reranking via Confidence-Ordered Denoising
-
UniRank formulates list-wise reranking as confidence-ordered iterative masked discrete diffusion with bidirectional slate context, unifying autoregressive and non-autoregressive paradigms as inference-time special cases while eliminating causal error propagation and slot independence assumptions.
- Your Embedding Model Is SMARTer Than You Think
-
Single-vector multimodal embedders naturally shape their non-pooling hidden states for cosine matching under pooled contrastive training; SMART leverages these frozen states via unweighted hybrid scoring (pooled + MaxSim) to boost fine-grained retrieval training-free and enables ultra-fast post-training that rivals dedicated multi-vector models.