AI 速览：Rimuru 提出共享页池与内存视图机制，统一管理 LLM 服务中异构张量的内存，并通过弹性专家卸载提升吞吐、降低首 token 延迟。
Modern LLMs strain device memory through large model weights and attention states, particularly in resource-constrained scenarios. Existing systems separately provision memory for model weights and attention states, imposing fixed fixed boundaries that cannot adapt to the changing memory demands of workloads. Under heavy workloads, insufficient state capacity limits request concurrency and reduces throughput; under light workloads, idle memory in one region cannot be reallocated to another, wasting device memory. Meanwhile, modern LLM architectures introduce heterogeneous tensors with diverse layouts and lifetimes, further complicating memory management. These problems motivate unified memory management for heterogeneous tensors. However, a shared pool alone does not make memory directly reusable across tensors with different allocation granularities and execution layouts. This paper presents Rimuru, a modern LLM serving system that unifies memory management through elastic expert offloading. Rimuru establishes a shared page pool to enables elastic memory sharing across heterogeneous tensors. To accommodate different allocation granularities, Rimuru introduces memory views that allow each tensor type to interpret the shared memory space using its own page size and page table. Furthermore, Rimuru combines huge-page alignment with a multi-level page allocator that splits and merges pages, enabling cross-type capacity reuse. A page-table-aware fused MoE kernel directly accesses non-contiguous expert weights, avoiding copy overhead. Building on these mechanisms, Rimuru employs a workload-aware adjuster that dynamically reallocates pages across tensor types according to runtime demands. Implemented on top of vLLM, Rimuru remains compatible with modern LLMs. Evaluations show that Rimuru achieves up to 19.03x baseline throughput and reduces TTFT by 98.1% on average.
Chenxiao Fan, Chongming Gao, Gangyi Zhang, Yaxin Gong, Leyang Shen, Jiakai Wang, Dong Wang, Yang Liu 等 12 位作者
Complex information seeking on the Web often requires collecting evidence from multiple documents and reasoning across them, as in multi-hop question answering. Multi-agent systems address such tasks by dividing work among specialized agents, where upstream agents produce intermediate answers and supporting evidence that downstream agents build upon. However, upstream answers vary in correctness, and most systems provide no quantitative signal about it, leaving the downstream agent to judge whether to trust or verify what it receives based solely on the textual evidence. In this paper, we propose MASCAL. Each agent learns to produce a correctness probability for its answer by jointly optimizing task performance and confidence calibration in reinforcement learning; the upstream agent passes this probability at handoff, giving the downstream agent a quantitative basis for its decisions, and the final answer also carries a calibrated probability, quantifying the reliability of the system output. To guide the downstream agent in using this signal during training, we introduce confidence-aware weighting, which increases the learning weight at positions responsive to the upstream probability, so that the agent learns more from decisions where it actively uses the signal. Across three collaboration architectures spanning Web retrieval and reasoning tasks, MASCAL improves task accuracy and calibration quality for both agents, and generalizes to deeper collaboration chains and different model scales. Intervention analysis confirms that upstream confidence affects downstream decision behavior, and the learned final confidence supports selective answering and targeted rechecking.
Modeling asynchronous event sequences is crucial in numerous real-world applications such as healthcare monitoring, financial transaction analysis, and so on. Traditional temporal point processes, including Hawkes Processes, often fail to capture complex dependencies due to their parametric limitations. While neural approaches like RNNs and Transformers have improved flexibility, they struggle with computational inefficiency, and attention saturation. In this paper, we introduce the Mamba Hawkes Process (MHP), the first framework to integrate selective state space model (Mamba) with temporal point processes. MHP leverages time-varying state transitions and input-dependent gating to efficiently encode event history and capture long-term dependencies with linear complexity. Importantly, we provide theoretical guarantees showing that MHP generalizes both classical multi-exponential Hawkes processes and exponential-decay gated RNNs, underscoring its expressive power and theoretical soundness. To address the inherent constraints of pure state space models in handling heterogeneous event interactions, we further develop Adaptive Mamba Hawkes Process (A-MHP) that incorporates two novel mechanisms: a Time-Scaling Mechanism that adaptively weights time intervals based on event type and history, and a Dual-Channel State Transition that adaptively processes event content and temporal dynamics for more refined state updates. Extensive experiments on synthetic and realworld datasets demonstrate that MHP and A-MHP consistently outperform state-of-the-art baselines in event prediction tasks, particularly in long-sequence scenarios. Our work establishes a scalable and theoretically grounded paradigm for event sequence modeling, with practical implications for predictive maintenance, anomaly detection, and dynamic system analysis.
We study conformal prediction under covariate shift with a known covariate density ratio, seeking prediction sets with minimum expected length under the target distribution. For every calibration sample size, we prove that the rule combining the inverse conditional-density score with randomized weighted conformal calibration minimizes this expected length among all rules valid over the full weighted-exchangeability class. We approximate the oracle using quantile and kernel histograms, including forest-weighted estimators, and select estimators and tuning parameters by resampling to estimate calibrated target length. On a bounded response domain, we bound excess length linearly in $L^1$ density error at fixed calibration size, without requiring a uniform positive lower bound on the true density. We also bound selection-pool and resampling errors. Under explicit growth conditions, the selected procedure approaches the same oracle while the calibration size remains fixed. Experiments on 36 known-density configurations and four public regression datasets compare coverage, prediction-set length, and runtime with existing methods. In targeted low-coverage and strong-shift comparisons, inverse-density selection yields shorter mean sets in most cases and lower mean runtime than the evaluated weighted baselines, with some differences in achieved coverage.
Zhiyuan Zhou, Andy Peng, Charles Xu, Qiyang Li, Jost Tobias Springenberg, Kevin Frans, Sergey Levine
Expressive continuous control policies, such as diffusion and flow models, form the backbone of recent advances in scaling imitation learning for simulated and real robot control. While they are known to scale stably in the supervised imitation learning setting, incorporating them into reinforcement learning (RL) pipelines for policy improvement has proven more difficult. It often requires specialized training objectives or backpropagating through denoising processes, which cause well-known issues with stability and affect scalability. In this paper we study the question of whether simple policy improvement schemes at test time alone, leaving stable supervised policy training intact, can be a competitive alternative which sidesteps these issues. To this end, we propose QGF (Q-Guided Flow), an RL algorithm that performs policy optimization entirely at test time. QGF works by pre-training both a reference flow policy (via a standard behavioral cloning objective) and a value function critic and, at test time, using the value gradient to guide the reference policy to generate higher-value actions without any additional policy learning. Empirically, QGF outperforms prior test-time RL methods on single-task and goal-conditioned offline RL benchmarks with high-dimensional action spaces, and is competitive with state-of-the-art training-time algorithms while being much cheaper to run. Moreover, it exhibits favorable scaling with model size by avoiding the instability of actor-critic training, offering a practical and effective alternative RL algorithm with expressive policies.
Qingyun Liu, Bo Yan, Yang Liu, Yuji Roh, Ekansh Sharma, Likang Yin, Emma Olowo, Yuxuan Li 等 19 位作者
User modeling in industrial recommender systems typically produces dense embeddings, which suffer from representational constraints due to fixed-dimensional vectors. As an emerging alternative for discrete user representation, LLM-generated natural-language user profiles often capture language semantics rather than sequential patterns in user behaviors. Meanwhile, Semantic ID (SID) is effective for item representation, yet existing discrete user representations remain largely confined to quantizing already-pooled user vectors, clustering historical watches, or predicting coarse topic clusters solely for candidate retrieval. We propose TokenMinds, an industrial system that extends the PLUM framework from SID-based item retrieval to generative sequential user modeling. Built on an encoder-decoder architecture adapted from pretrained LLMs, TokenMinds introduces a dual-output design: the encoder outputs a dense user embedding capturing a holistic user state, while the decoder generates discrete SID-based user tokens representing multi-faceted future interests. Grounded in item semantics, these tokens preserve distinct interests that a single pooled vector would average away. Additionally, a shared SID vocabulary across long- and short-form videos unifies cross-scenario behaviors into a single model, cutting upstream training and serving compute by 50% and 31%. Furthermore, leveraging the text vocabulary inherited from the pretrained LLM, TokenMinds interleaves textual search queries directly with watch histories to capture explicit user intent. We validate TokenMinds through offline evaluations and live A/B experiments across YouTube's primary ranking surfaces, serving billions of users asynchronously: SID-based user tokens are effective on their own, yield additive gains alongside dense embeddings, and together outperform the previous-generation, non-LLM user model in production.
Yang Liu, Yunjiang Jiang, Ayush Agarwal, Yihan Wu, Haoran Liu, Bi Xue
Scaling large recommendation systems requires advancing three major frontiers: processing longer user histories, expanding candidate sets, and increasing model capacity. While promising, transformers' computational cost scales quadratically with the user sequence length and linearly with the number of candidates. This trade-off makes it prohibitively expensive to expand candidate sets or increase sequence length at inference, despite the significant performance improvements. We introduce \textbf{LIME}, a novel architecture that resolves this trade-off. Through two key innovations, LIME fundamentally reduces computational complexity. First, low-rank ``link embeddings" enable pre-computation of attention weights by decoupling user and candidate interactions, making the inference cost nearly independent of candidate set size. Second, a linear attention mechanism, \textbf{LIME-XOR}, reduces the complexity with respect to user sequence length from quadratic ($O(N^2)$) to linear ($O(N)$). Experiments on public and industrial datasets show LIME achieves near-parity with state-of-the-art transformers but with a 10$\times$ inference speedup on large candidate sets or long sequence lengths. When tested on a major recommendation platform, LIME improved user engagement while maintaining minimal inference costs with respect to candidate set size and user history length, establishing a new paradigm for efficient and expressive recommendation systems.
Xiangyu Liu, Zeting Yan, Zhitong Yin, Boyang Li, Xi Zhang
Background and Objective: Predicting regional blood-oxygen-level-dependent (BOLD) activity from EEG provides a computational approach to studying electrophysiological-hemodynamic associations. However, changes in input representation, readout architecture, and ensembling can obscure the source of a prediction gain. We investigate the contribution of target-relative spectral estimation under matched readout conditions. Methods: MEL converts each preceding 16-second EEG history into calibrated lag-channel-frequency measurements. Four equal-dimensional representations are compared using matched Ridge and multilayer perceptron (MLP) readouts. The retrospective analysis includes 29 resting-state scans from 22 participants, participant-paired uncertainty estimates, aggregation and correspondence controls, leave-one-participant-out regression, and refitted models on an external Oddball cohort. Results: Morlet-Ridge achieved a mean scan-wise correlation of 0.508 across six regional targets and the global signal, compared with 0.432 for canonical Fourier features. The participant-paired mean difference was 0.069 (95% bootstrap confidence interval, 0.037-0.101), with improvements in 19 of 22 participants. The Morlet MLP prediction ensemble reached 0.491. Shared coordinate permutations preserved prediction after refitting, whereas sample-wise permutations degraded it. Canonical Fourier regression reached 0.377 under participant exclusion; external unseen-participant correlation was 0.026. Conclusions: Spectral estimation materially affected regional BOLD prediction without requiring additional nonlinear readout capacity. MEL provides an explicit measurement interface for controlled comparisons, but weak external results constrain its generalizability. Confirmation requires untouched participant-separated data and split-aware preprocessing.
Edric Tam, Qingyang Liu, Barbara Engelhardt, Ray Bai
Broad classes of deep generative models (DGMs) operate by learning to map a simple source distribution, such as a standard Gaussian, to a complex high-dimensional target distribution. Despite their success, DGMs often empirically struggle to accurately represent distributions with heavy tails or well-separated modes. In this work, we show that these difficulties reflect intrinsic properties of the underlying transport problem rather than particular neural network architectures or training procedures. We establish quantitative lower bounds on the gradients and Jacobians of pushforward mappings that transport log-concave source distributions to such targets. For heavy-tailed target distributions, the required gradient magnitude grows as the target develops increasingly large moments. For multimodal target distributions, the low-probability regions between well-separated modes similarly force large gradients in the transport map. We discuss how such large derivatives relate to trade-offs between expressivity and numerical stability. Our results hold under substantially weaker regularity conditions than previous analyses and remain stable under perturbations of the target distribution. Finally, we show that these obstructions can be mitigated when the goal is approximation of a given density rather than generative modeling. Specifically, we propose simple classes of bijective reparametrizations that can alleviate the difficult geometric structure and reduce the gradient burden placed on the learned generative map. Numerical experiments support our theoretical analysis.
Peter Kairouz, H. Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles 等 59 位作者
AI 速览：一篇综述性论文，梳理联邦学习的研究进展并系统列出开放问题与挑战。
Federated learning (FL) is a machine learning setting where many clients (e.g. mobile devices or whole organizations) collaboratively train a model under the orchestration of a central server (e.g. service provider), while keeping the training data decentralized. FL embodies the principles of focused data collection and minimization, and can mitigate many of the systemic privacy risks and costs resulting from traditional, centralized machine learning and data science approaches. Motivated by the explosive growth in FL research, this paper discusses recent advances and presents an extensive collection of open problems and challenges.