Fast Spatial Memory with Elastic Test-Time Training

Large Chunk Test-Time Training (LaCT) has shown strong performance on long-context 3D reconstruction, but its fully plastic inference-time updates remain vulnerable to catastrophic forgetting and overfitting. As a result, LaCT is typically instantiated with a single large chunk spanning the full input sequence, falling short of the broader goal of handling arbitrarily long sequences in a single pass. We propose Elastic Test-Time Training inspired by elastic weight consolidation, that stabilizes LaCT fast-weight updates with a Fisher-weighted elastic prior around a maintained anchor state. The anchor evolves as an exponential moving average of past fast weights to balance stability and plasticity. Based on this updated architecture, we introduce Fast Spatial Memory (FSM), an efficient and scalable model for 4D reconstruction that learns spatiotemporal representations from long observation sequences and renders novel view-time combinations. We pre-trained FSM on large-scale curated 3D/4D data to capture the dynamics and semantics of complex spatial environments. Extensive experiments show that FSM supports fast adaptation over long sequences and delivers high-quality 3D/4D reconstruction with smaller chunks and mitigating the camera-interpolation shortcut. Overall, we hope to advance LaCT beyond the bounded single-chunk setting toward robust multi-chunk adaptation, a necessary step for generalization to genuinely longer sequences, while substantially alleviating the activation-memory bottleneck.

DCD: Domain-Oriented Design for Controlled Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) is widely used to ground large language models in external knowledge sources. However, when applied to heterogeneous corpora and multi-step queries, Naive RAG pipelines often degrade in quality due to flat knowledge representations and the absence of explicit workflows. In this work, we introduce DCD (Domain-Collection-Document), a domain-oriented design to structure knowledge and control query processing in RAG systems without modifying the underlying language model. The proposed approach relies on a hierarchical decomposition of the information space and multi-stage routing based on structured model outputs, enabling progressive restriction of both retrieval and generation scopes. The architecture is complemented by smart chunking, hybrid retrieval, and integrated validation and generation guardrail mechanisms. We describe the DCD architecture and workflow and discuss evaluation results on synthetic evaluation dataset, highlighting their impact on robustness, factual accuracy, and answer relevance in applied RAG scenarios.

AV-SQL: Decomposing Complex Text-to-SQL Queries with Agentic Views

Text-to-SQL is the task of translating natural language queries into executable SQL for a given database, enabling non-expert users to access structured data without writing SQL manually. Despite rapid advances driven by large language models (LLMs), existing approaches still struggle with complex queries in real-world settings, where database schemas are large and questions require multi-step reasoning over many interrelated tables. In such cases, providing the full schema often exceeds the context window, while one-shot generation frequently produces non-executable SQL due to syntax errors and incorrect schema linking. To address these challenges, we introduce AV-SQL, a framework that decomposes complex Text-to-SQL into a pipeline of specialized LLM agents. Central to AV-SQL is the concept of agentic views: agent-generated Common Table Expressions (CTEs) that encapsulate intermediate query logic and filter relevant schema elements from large schemas. AV-SQL operates in three stages: (1) a rewriter agent compresses and clarifies the input query; (2) a view generator agent processes schema chunks to produce agentic views; and (3) a planner, generator, and revisor agent collaboratively compose these views into the final SQL query. Extensive experiments show that AV-SQL achieves 70.38% execution accuracy on the challenging Spider 2.0 benchmark, outperforming state-of-the-art baselines, while remaining competitive on standard datasets with 85.59% on Spider, 72.16% on BIRD and 63.78% on KaggleDBQA. Our source code is available at https://github.com/pminhtam/AV-SQL.

DTCRS: Dynamic Tree Construction for Recursive Summarization

Retrieval-Augmented Generation (RAG) mitigates the hallucination problem of Large Language Models (LLMs) by incorporating external knowledge. Recursive summarization constructs a hierarchical summary tree by clustering text chunks, integrating information from multiple parts of a document to provide evidence for abstractive questions involving multi-step reasoning. However, summary trees often contain a large number of redundant summary nodes, which not only increase construction time but may also negatively impact question answering. Moreover, recursive summarization is not suitable for all types of questions. We introduce DTCRS, a method that dynamically generates summary trees based on document structure and query semantics. DTCRS determines whether a summary tree is necessary by analyzing the question type. It then decomposes the question and uses the embeddings of sub-questions as initial cluster centers, reducing redundant summaries while improving the relevance between summaries and the question. Our approach significantly reduces summary tree construction time and achieves substantial improvements across three QA tasks. Additionally, we investigate the applicability of recursive summarization to different question types, providing valuable insights for future research.

HybridKV: Hybrid KV Cache Compression for Efficient Multimodal Large Language Model Inference

Multimodal Large Language Models (MLLMs) have advanced unified reasoning over text, images, and videos, but their inference is hindered by the rapid growth of key-value (KV) caches. Each visual input expands into thousands of tokens, causing caches to scale linearly with context length and remain resident in GPU memory throughout decoding, which leads to prohibitive memory overhead and latency even on high-end GPUs. A common solution is to compress caches under a fixed allocated budget at different granularities: token-level uniformly discards less important tokens, layer-level varies retention across layers, and head-level redistributes budgets across heads. Yet these approaches stop at allocation and overlook the heterogeneous behaviors of attention heads that require distinct compression strategies. We propose HybridKV, a hybrid KV cache compression framework that integrates complementary strategies in three stages: heads are first classified into static or dynamic types using text-centric attention; then a top-down budget allocation scheme hierarchically assigns KV budgets; finally, static heads are compressed by text-prior pruning and dynamic heads by chunk-wise retrieval. Experiments on 11 multimodal benchmarks with Qwen2.5-VL-7B show that HybridKV reduces KV cache memory by up to $7.9\times$ and achieves $1.52\times$ faster decoding, with almost no performance drop or even higher relative to the full-cache MLLM.

In-Place Test-Time Training

The static ``train then deploy" paradigm fundamentally limits Large Language Models (LLMs) from dynamically adapting their weights in response to continuous streams of new information inherent in real-world tasks. Test-Time Training (TTT) offers a compelling alternative by updating a subset of model parameters (fast weights) at inference time, yet its potential in the current LLM ecosystem is hindered by critical barriers including architectural incompatibility, computational inefficiency and misaligned fast weight objectives for language modeling. In this work, we introduce In-Place Test-Time Training (In-Place TTT), a framework that seamlessly endows LLMs with Test-Time Training ability. In-Place TTT treats the final projection matrix of the ubiquitous MLP blocks as its adaptable fast weights, enabling a ``drop-in" enhancement for LLMs without costly retraining from scratch. Furthermore, we replace TTT's generic reconstruction objective with a tailored, theoretically-grounded objective explicitly aligned with the Next-Token-Prediction task governing autoregressive language modeling. This principled objective, combined with an efficient chunk-wise update mechanism, results in a highly scalable algorithm compatible with context parallelism. Extensive experiments validate our framework's effectiveness: as an in-place enhancement, it enables a 4B-parameter model to achieve superior performance on tasks with contexts up to 128k, and when pretrained from scratch, it consistently outperforms competitive TTT-related approaches. Ablation study results further provide deeper insights on our design choices. Collectively, our results establish In-Place TTT as a promising step towards a paradigm of continual learning in LLMs.

HiPolicy: Hierarchical Multi-Frequency Action Chunking for Policy Learning

Robotic imitation learning faces a fundamental trade-off between modeling long-horizon dependencies and enabling fine-grained closed-loop control. Existing fixed-frequency action chunking approaches struggle to achieve both. Building on this insight, we propose HiPolicy, a hierarchical multi-frequency action chunking framework that jointly predicts action sequences at different frequencies to capture both coarse high-level plans and precise reactive motions. We extract and fuse hierarchical features from history observations aligned to each frequency for multi-frequency chunk generation, and introduce an entropy-guided execution mechanism that adaptively balances long-horizon planning with fine-grained control based on action uncertainty. Experiments on diverse simulated benchmarks and real-world manipulation tasks show that HiPolicy can be seamlessly integrated into existing 2D and 3D generative policies, delivering consistent improvements in performance while significantly enhancing execution efficiency.

Offline RL for Adaptive Policy Retrieval in Prior Authorization

Prior authorization (PA) requires interpretation of complex and fragmented coverage policies, yet existing retrieval-augmented systems rely on static top-$K$ strategies with fixed numbers of retrieved sections. Such fixed retrieval can be inefficient and gather irrelevant or insufficient information. We model policy retrieval for PA as a sequential decision-making problem, formulating adaptive retrieval as a Markov Decision Process (MDP). In our system, an agent iteratively selects policy chunks from a top-$K$ candidate set or chooses to stop and issue a decision. The reward balances decision correctness against retrieval cost, capturing the trade-off between accuracy and efficiency. We train policies using Conservative Q-Learning (CQL), Implicit Q-Learning (IQL), and Direct Preference Optimization (DPO) in an offline RL setting on logged trajectories generated from baseline retrieval strategies over synthetic PA requests derived from publicly available CMS coverage data. On a corpus of 186 policy chunks spanning 10 CMS procedures, CQL achieves 92% decision accuracy (+30 percentage points over the best fixed-$K$ baseline) via exhaustive retrieval, while IQL matches the best baseline accuracy using 44% fewer retrieval steps and achieves the only positive episodic return among all policies. Transition-level DPO matches CQL's 92% accuracy while using 47% fewer retrieval steps (10.6 vs. 20.0), occupying a "selective-accurate" region on the Pareto frontier that dominates both CQL and BC. A behavioral cloning baseline matches CQL, confirming that advantage-weighted or preference-based policy extraction is needed to learn selective retrieval. Lambda ablation over step costs $λ\in \{0.05, 0.1, 0.2\}$ reveals a clear accuracy-efficiency inflection: only at $λ= 0.2$ does CQL transition from exhaustive to selective retrieval.

MemMachine: A Ground-Truth-Preserving Memory System for Personalized AI Agents

Large Language Model (LLM) agents require persistent memory to maintain personalization, factual continuity, and long-horizon reasoning, yet standard context-window and retrieval-augmented generation (RAG) pipelines degrade over multi-session interactions. We present MemMachine, an open-source memory system that integrates short-term, long-term episodic, and profile memory within a ground-truth-preserving architecture that stores entire conversational episodes and reduces lossy LLM-based extraction. MemMachine uses contextualized retrieval that expands nucleus matches with surrounding context, improving recall when relevant evidence spans multiple dialogue turns. Across benchmarks, MemMachine achieves strong accuracy-efficiency tradeoffs: on LoCoMo it reaches 0.9169 using gpt4.1-mini; on LongMemEvalS (ICLR 2025), a six-dimension ablation yields 93.0 percent accuracy, with retrieval-stage optimizations -- retrieval depth tuning (+4.2 percent), context formatting (+2.0 percent), search prompt design (+1.8 percent), and query bias correction (+1.4 percent) -- outperforming ingestion-stage gains such as sentence chunking (+0.8 percent). GPT-5-mini exceeds GPT-5 by 2.6 percent when paired with optimized prompts, making it the most cost-efficient setup. Compared to Mem0, MemMachine uses roughly 80 percent fewer input tokens under matched conditions. A companion Retrieval Agent adaptively routes queries among direct retrieval, parallel decomposition, or iterative chain-of-query strategies, achieving 93.2 percent on HotpotQA-hard and 92.6 percent on WikiMultiHop under randomized-noise conditions. These results show that preserving episodic ground truth while layering adaptive retrieval yields robust, efficient long-term memory for personalized LLM agents.

CAWN: Continuous Acoustic Wave Networks for Autoregressive Language Modeling

Modern Large Language Models (LLMs) rely on Transformer self-attention, which scales quadratically with sequence length. Recent linear-time alternatives, like State Space Models (SSMs), often suffer from signal degradation over extended contexts. We introduce the Continuous Acoustic Wave Network (CAWN), a fully continuous sequence-mixing architecture. Instead of discrete matrix-based attention, CAWN projects hidden states into multi-headed complex-domain phasors, achieving sequence mixing through a causal, $O(L)$ Phase Accumulation mechanism. To prevent signal degradation over ultra-long contexts, we introduce a dual-gated Selective Phase Resonance mechanism incorporating Frequency-Dependent Retention, Hard-Threshold Gating via Straight-Through Estimation, and a Temporal Syntax Cache to capture short-term local dependencies. We also replace standard dense linear projections with Depth-wise Harmonic Convolutions for optimal spatial frequency mixing, augmented by Block Attention Residuals for depth-wise state routing. Scaled to a 150M-parameter model, CAWN utilizes custom Triton kernels for hardware-efficient, true-complex phase accumulation in float32. Trained via a continuous streaming loop on a 100-Billion-token corpus, the prototype is evaluated at a 5-Billion-token milestone. Empirical evaluations via a Targeted Semantic Retrieval protocol demonstrate robust vocabulary acquisition and extended explicitly learned contextual denoising. By leveraging $O(1)$ state-passing via chunked prefill, the model retrieves targeted information across 2,000,000 tokens while strictly plateauing at 8.72 GB of Peak VRAM, empirically overcoming the $O(L^2)$ context memory wall.