With the rapid advancement of human science and technology, problems in industrial scenarios are becoming increasingly challenging, bringing significant challenges to traditional algorithm design. Automated algorithm design with LLMs emerges as a promising solution, but the currently adopted black-box modeling deprives LLMs of any awareness of the intrinsic mechanism of the target problem, leading to hallucinated designs. In this paper, we introduce Evolutionary Stagewise Algorithm Design (EvoStage), a novel evolutionary paradigm that bridges the gap between the rigorous demands of industrial-scale algorithm design and the LLM-based algorithm design methods. Drawing inspiration from CoT, EvoStage decomposes the algorithm design process into sequential, manageable stages and integrates real-time intermediate feedback to iteratively refine algorithm design directions. To further reduce the algorithm design space and avoid falling into local optima, we introduce a multi-agent system and a "global-local perspective" mechanism. We apply EvoStage to the design of two types of common optimizers: designing parameter configuration schedules of the Adam optimizer for chip placement, and designing acquisition functions of Bayesian optimization for black-box optimization. Experimental results across open-source benchmarks demonstrate that EvoStage outperforms human-expert designs and existing LLM-based methods within only a couple of evolution steps, even achieving the historically state-of-the-art half-perimeter wire-length results on every tested chip case. Furthermore, when deployed on a commercial-grade 3D chip placement tool, EvoStage significantly surpasses the original performance metrics, achieving record-breaking efficiency. We hope EvoStage can significantly advance automated algorithm design in the real world, helping elevate human productivity.

We study dynamic regret minimization in non-stationary online learning, with a primary focus on follow-the-regularized-leader (FTRL) methods. FTRL is important for curved losses and for understanding adaptive optimizers such as Adam, yet existing dynamic regret analyses are less explored for FTRL. To address this, we build on the discounted-to-dynamic reduction and present a modular way to obtain dynamic regret bounds of FTRL-related problems. Specifically, we focus on two representative curved losses: linear regression and logistic regression. Our method not only simplifies existing proofs for the optimal dynamic regret of online linear regression, but also yields new dynamic regret guarantees for online logistic regression. Beyond online convex optimization, we apply the reduction to analyze the Adam optimizers, obtaining optimal convergence rates in stochastic, non-convex, and non-smooth settings. The reduction also enables a more detailed treatment of Adam with two discount parameters (β_1,β_2), leading to new results for both clipped and clip-free variants of Adam optimizers.

While Mixture-of-Experts (MoE) architectures substantially bolster the expressive power of large-language models, their prohibitive memory footprint severely impedes the practical deployment on resource-constrained edge devices, especially when model behavior must be preserved without relying on lossy quantization. In this paper, we present ZipMoE, an efficient and semantically lossless on-device MoE serving system. ZipMoE exploits the synergy between the hardware properties of edge devices and the statistical redundancy inherent to MoE parameters via a caching-scheduling co-design with provable performance guarantee. Fundamentally, our design shifts the paradigm of on-device MoE inference from an I/O-bound bottleneck to a compute-centric workflow that enables efficient parallelization. We implement a prototype of ZipMoE and conduct extensive experiments on representative edge computing platforms using popular open-source MoE models and real-world workloads. Our evaluation reveals that ZipMoE achieves up to 72.77% inference latency reduction and up to 6.76× higher throughput than the state-of-the-art systems.

Spiking Neural Networks (SNNs) have garnered increasing attention as one of bio-inspired models due to their great potential in neuromorphic computing and sparse computation. Many practical algorithms and techniques have been developed; however, theoretical understandings of the generalization, that is, the extent to which SNNs perform well on unseen data, are far from clear. Recent advances disclosed an excitation-dependent and architecture-related generalization bound such that the Rademacher complexity of SNNs with stochastic firing can be upper bounded by an exponential function relative to the excitation probability and the architecture depth. In this paper, we theoretically investigate the generalization bounds of SNNs with several integration-and-fire schemes via Rademacher complexity. We recognize that the empirical Rademacher complexity of SNNs is close to the SNN configurations, which is exponential to the network depth and the maximum time duration of received spike sequences, superlinear and subquadratic to the network width, polynomial to the parameter norm, inverse-linear to the number of training samples, and independent of the computations within spiking neurons, achieving a more precise rate than conventional studies. Our theoretical results may support the scope of SNN theories and shed some insight into the development of SNNs.

Speculative decoding, in which a lightweight draft model first generates a draft sequence that is then verified in parallel by the target model, has become a prevalent paradigm for accelerating large language model inference. Recent work such as DFlash further boosts drafting efficiency by leveraging diffusion drafters, whose parallel denoising mechanism enables draft generation in a single forward pass. In this work, we uncover a central pitfall of diffusion drafters: bidirectional attention is a double-edged sword. On one hand, it endows the model with parallel generation and global contextual modeling capabilities; on the other hand, this inherent global dependency introduces high variance at both the domain-level and the token-level: acceptance rates fluctuate substantially across different domains, and draft token quality also varies heterogeneously at different token positions. To tackle this issue, we propose AdaFlash framework, comprising two components: (i) an on-policy distillation (OPD) algorithm with reverse-KL divergence tailored for diffusion drafters, bringing stable convergence and effectively reducing domain-level variance; and (ii) an adaptive length head that dynamically adjusts the candidate sequence length on the fly, substantially lowering the verification cost of the target model and handling token-level variance. Experiments demonstrate that AdaFlash consistently improves speedup rate during deployment, with especially significant gains in high-concurrency scenarios, achieving up to approximately 66

Chip placement is a critical step in physical design. While reinforcement learning (RL)-based methods have recently emerged, their training primarily focuses on wirelength optimization, and therefore often fail to achieve expert-quality layouts. We identify the reward design as the primary cause for the performance gap with experts, and instead of formalizing intricate processes, we circumvent this by directly learning from expert layouts to derive a reward model. Our approach starts from the final expert layouts to infer step-by-step expert trajectories. Using these trajectories as demonstrations or preferences, we train a model that captures the latent implicit rewards in expert results. Experiments show that our framework can efficiently learn from even a single design and generalize well to unseen cases.

Machine learning under limited computational resources has gained increasing attention recently. A common yet challenging scenario is managing multiple time-constrained learning tasks with budgeted computational resources, known as Computational Resource Efficient Learning (CoRE-Learning). To this end, a recently proposed framework, Learning with Adaptive Resource Allocation (LARA), offers a preliminary approach. In this paper, we point out the limitations of LARA, including its reliance on interpolation-based extrapolation methods, the need for a fixed exploration phase, and the use of high-frequency re-estimation and reallocation strategies. To address these issues, we propose Look-ahead and immediate Resource Allocation (LaiRA). Our approach incorporates an efficient Dynamic Kalman Filtering (DKF) for look-ahead feasibility check with limited data and a weight-based online estimator for immediate performance evaluation. For resource allocation, LaiRA constructs an Upper Confidence Bound (UCB) to enable adaptive exploration and introduces an adaptive time-slicing method to reduce task switching costs. Empirical studies validate the effectiveness of our approach.

The paradigm aims to enable users to leverage numerous existing high-performing models instead of building machine learning models from scratch. A learnware consists of a submitted model together with a derived from the developer’s training data. As the key component, a specification should characterize the capabilities of the model, enabling it to be adequately identified and reused, while preserving the developer's original data. In this paper, we present the first formal study of the risks that arise when a specification is attached to a model, as opposed to releasing the model alone. We develop a game-theoretic framework and, by combining variational inference with geometry analysis, provide quantitative estimates of the resulting risk of specification. Our analysis provides theoretical guarantees on the data protection ability for the commonly adopted RKME specification. Finally, we prove that with a properly chosen size of specification, releasing the specification alongside the model introduces almost no additional risk of exposing the raw data, while still retaining sufficient information for effective learnware identification.

“Learnware = Model + Specification”. The learnware paradigm aims to help users reuse existing high-performing models instead of training models from scratch, where the specification characterizes a trained model's capability and utility without exposing raw data. Numerous learnwares are hosted by a learnware dock system, which identifies helpful learnwares according to user task requirements and returns suitable models for reuse. In practice, learnwares that exactly match a user's feature space are often rare, while models from heterogeneous feature spaces, or even heterogeneous label spaces, may still be reusable. The central challenge is to characterize and compare the reusability of such heterogeneous models within a unified representation space. This paper finds that label information, especially model outputs, is crucial but previously insufficiently used. We exploit model outputs to evolve specifications into a unified space, thereby constructing a unified capability representation of heterogeneous models that characterizes their predictive behavior across diverse feature and label spaces. To instantiate this idea, we extend the specification implementation to encode model-output behavior more effectively and prevent it from being overwhelmed by high-dimensional features. Based on the unified capability representation, we identify learnwares by matching the conditional distributions induced by model outputs with the true distributions of user tasks, enabling learnwares to be reused beyond their original purposes. Experiments show that the proposed method can effectively identify and assemble learnwares from diverse feature and label spaces, even when no submitted learnware is explicitly tailored to the user task.

The *Learnware* paradigm aims to help users solve machine learning tasks by leveraging existing well-trained models rather than starting from scratch. A learnware comprises a submitted model paired with a *specification* sketching its capabilities. For an open platform with continuously uploaded models, these specifications are essential to enabling users to identify helpful models, eliminating the requirement for prohibitively costly per-model evaluations. In previous research, specifications based on privacy-preserving reduced sets succeed in enabling learnware identification through distribution matching, but suffer from high sample complexity for learnwares from high-dimensional, unstructured data like images or text. In this paper, we formalize **Pa**rameter **Ve**ctor (PAVE) specification for learnware identification, which utilizes the changes in pre-trained model parameters to inherently encode the model capability and task requirements, offering an effective solution for these learnwares. Theoretically, from the neural tangent kernel perspective, we establish a tight connection between PAVE and prior specifications, providing a theoretical explanation for their shared underlying principles. We further approximate the parameter vector in a low-rank space and analyze the approximation error bound, highly reducing the computational and storage overhead. Extensive empirical studies demonstrate that PAVE specification excels at identifying CV and NLP learnwares for reuse on given user tasks, and succeeds in identifying helpful learnwares from open learnware repository with corrupted model quality for the first time. Reusing identified learnware to solve user tasks can even outperform user-fine-tuned pre-trained models in data-limited scenarios.

Online eXp-concave Optimization (OXO) is a fundamental problem in online learning. The standard algorithm, Online Newton Step (ONS), balances statistical optimality and computational practicality, guaranteeing an optimal regret of $O(d \log T)$, where $d$ is the dimension and $T$ is the time horizon. ONS faces a computational bottleneck due to the Mahalanobis projections at each round. This step costs $Ω(d^ω)$ arithmetic operations for bounded domains, even for the unit ball, where $ω\in (2,3]$ is the matrix-multiplication exponent. As a result, the total runtime can reach $\tilde{O}(d^ωT)$, particularly when iterates frequently oscillate near the domain boundary. For Stochastic eXp-concave Optimization (SXO), computational cost is also a challenge. Deploying ONS with online-to-batch conversion for SXO requires $T = \tilde{O}(d/ε)$ rounds to achieve an excess risk of $ε$, and thereby necessitates an $\tilde{O}(d^{ω+1}/ε)$ runtime. A COLT'13 open problem posed by Koren [2013] asks for an SXO algorithm with runtime less than $\tilde{O}(d^{ω+1}/ε)$. This paper proposes a simple variant of ONS, LightONS, which reduces the total runtime to $O(d^2 T + d^ω\sqrt{T \log T})$ while preserving the optimal $O(d \log T)$ regret. LightONS implies an SXO method with runtime $\tilde{O}(d^3/ε)$, thereby answering the open problem. Importantly, LightONS preserves the elegant structure of ONS by leveraging domain-conversion techniques from parameter-free online learning to introduce a hysteresis mechanism that delays expensive Mahalanobis projections until necessary. This design enables LightONS to serve as an efficient plug-in replacement of ONS in broader scenarios, even beyond regret minimization, including gradient-norm adaptive regret, parametric stochastic bandits, and memory-efficient online learning.

Non-stationary parametric bandits have attracted much attention recently. There are three principled ways to deal with non-stationarity, including sliding-window, weighted, and restart strategies. As many non-stationary environments exhibit gradual drifting patterns, the weighted strategy is commonly adopted in real-world applications. However, previous theoretical studies show that its analysis is more involved and the algorithms are either computationally less efficient or statistically suboptimal. This paper revisits the weighted strategy for non-stationary parametric bandits. In linear bandits (LB), we discover that this undesirable feature is due to an inadequate regret analysis, which results in an overly complex algorithm design. We propose a refined analysis framework, which simplifies the derivation and importantly produces a simpler weight-based algorithm that is as efficient as window/restart-based algorithms while retaining the same regret as previous studies. Furthermore, our new framework can be used to improve regret bounds of other parametric bandits, including Generalized Linear Bandits (GLB) and Self-Concordant Bandits (SCB). For example, we develop a simple weighted GLB algorithm with an $\widetilde{O}(k_\mu^{\frac{5}{4}} c_\mu^{-\frac{3}{4}} d^{\frac{3}{4}} P_T^{\frac{1}{4}}T^{\frac{3}{4}})$ regret, improving the $\widetilde{O}(k_\mu^{2} c_\mu^{-1}d^{\frac{9}{10}} P_T^{\frac{1}{5}}T^{\frac{4}{5}})$ bound in prior work, where $k_\mu$ and $c_\mu$ characterize the reward model's nonlinearity, $P_T$ measures the non-stationarity, $d$ and $T$ denote the dimension and time horizon.

Chip placement determines the locations of electronic components on a chip layout, which directly impacts performance, power, and area (PPA) metrics, and thus is a critical step in electronic design automation (EDA). As modern chips scale to accommodate millions of components, manual placement by human experts becomes infeasible, necessitating the use of automated algorithms. Recently, reinforcement learning (RL) has emerged as a promising approach for automating macro placement, owing to its high optimization efficiency and potential for generalization. Despite their promise, existing RL-based methods often neglect the value of expert knowledge accumulated through years of engineering practice. They tend to optimize oversimplified proxy objectives, resulting in suboptimal placements that deviate significantly from expert-designed solutions. To bridge this gap, we propose a novel RL-based placement framework that integrates EDA domain expertise from two complementary perspectives: (1) \textit{Expert Knowledge Injection}: Incorporating well-established placement knowledge, such as dataflow guidance, periphery bias, macro grouping, and I/O keepout constraints, to guide the learning process toward human-level solutions. (2) \textit{Expert Workflow Imitation}: Emulating the post-refinement process of human experts (i.e., updating the design iteratively based on backend PPA feedback) to progressively optimize timing metrics by employing preference optimization. Experiments on the ICCAD 2015 and OpenROAD benchmarks demonstrate that our method achieves substantial improvements in PPA metrics~(e.g., 12.84\% in total negative slack and 20.79\% in worst negative slack compared to the runner-up method on average), outperforming advanced analytical, black-box optimization, and RL-based methods.

Output-length prediction is important for efficient LLM serving, as it directly affects batching, memory reservation, and scheduling. For prompt-only length prediction, most existing methods use a one-shot sampled length as the label, implicitly treating each prompt as if it had one true target length. We show that this is unreliable: even under a fixed model and decoding setup, the same prompt induces a prompt-conditioned output length distribution, not a deterministic scalar, and this distribution is consistent with heavy-tailed behavior. Motivated by this, we cast length prediction as robust estimation from heavy-tailed prompt-conditioned length distributions. We propose prompt-conditioned length distribution (ProD) methods, which construct training targets from multiple independent generations of the same prompt. Two variants are developed to reuse the served LLM's hidden states: , which uses a median-based target for robust point prediction, and ProD-D, which uses a distributional target that preserves prompt-conditioned uncertainty. We provide theoretical justifications by analyzing the estimation error under a surrogate model. Experiments across diverse scenarios show consistent gains in prediction quality.

In this paper we first study the problem of generalized linear bandit (GLB) under heavy-tailed noise. The characteristics of heavy-tailed distributions are widely observed in real-world applications such as personalized recommendation, financial markets, and medical treatments. Based on the online mirror descent (OMD) method, we propose an algorithm EHM that extends the adaptive Huber loss method (Wang et al., 2025) with one-pass update (𝒪(1) computational complexity with respect to current round t and the time horizon T), which simultaneously achieves an almost optimal regret of 𝒪(T^1/1+ε) where T is the time horizon. In addition, by utilizing a special property of some link function (Sawarni et al., 2025), our algorithm eliminates the need to know a commonly used parameter. Next, we study the GLB problem under the case when contextual characteristic becomes piecewise constant, and we slightly revised former algorithm to obtain the PGLB-EHM algorithm. After theoretical analysis, we prove that the regret upper bound order stays the same. Furthermore, we look deeper into a special case of nonlinear bandit (NB) and present the NB-EHM algorithm with bisection method and special restriction. Eventually we utilize the affine lifting approach and show that the general NB problem can be applied with NB-EHM to achieve a sublinear regret bound.

The learnware paradigm aims to help users solve new tasks by reusing existing models rather than starting from scratch. A learnware consists of a model and the specification describing its capabilities. Numerous learnwares are accommodated by the learnware dock system. When users solve tasks with the system, learnwares that fully match the user task are often scarce or unavailable. This paper focuses on tabular classification tasks and explores reusing learnwares for new user tasks with significantly different feature and label spaces, leveraging the potential of numerous existing specialized tabular models developed for various tasks. Under the learnware paradigm, we find that tabular learnwares that seem semantically irrelevant can sometimes be beneficial for new user tasks. The proposed method relies solely on model-predicted probabilities and does not require gradient information, making it applicable to a wide range of tabular models. Experiments suggest that tabular learnwares can be reused beyond their original purpose across heterogeneous tasks.

In machine learning, a critical class of decision-related problems concerns preventing predicted undesirable outcomes, referred to as the avoiding undesired future (AUF) problem. To address this, the rehearsal learning framework has been proposed to model influence relations for effective decisions. However, existing rehearsal methods rely on restrictive parametric assumptions such as linear systems or additive noise, limiting their practical applicability. In this paper, we propose the first non-parametric rehearsal learning approach for AUF without assuming specific functional forms of data generation processes. Specifically, we use kernel machinery to reformulate the AUF objective into a unified representation that disentangles desirability modeling from action-induced distributional changes. To handle the discontinuity of desirability indicator, we present a smooth Probit surrogate and provide an approximation error bound. Meanwhile, we capture the action-induced changes via conditional mean embeddings, and develop a kernel ridge regression based nested estimator for AUF objective with consistency guarantees. Such a formulation naturally accommodates nonlinear systems and non-additive noise, and empirical results on synthetic and real-data-derived semi-synthetic benchmarks demonstrate the effectiveness and flexibility of our approach.

Isolation Forest (iForest) stands out as a widely-used unsupervised anomaly detector, primarily owing to its remarkable runtime efficiency and superior performance in large-scale tasks. Despite its widespread adoption, a theoretical foundation explaining iForest's success remains unclear. This paper focuses on the inductive bias of iForest, which theoretically elucidates under what circumstances and to what extent iForest works well. The key is to formulate the growth process of iForest, where the split dimensions and split values are randomly selected. We model the growth process of iForest as a random walk, enabling us to derive the expected depth function, which is the outcome of iForest, using transition probabilities. The case studies reveal key inductive biases: iForest exhibits lower sensitivity to central anomalies while demonstrating greater parameter adaptability compared to k-Nearest Neighbor. Our study provides a theoretical understanding of the effectiveness of iForest and establishes a foundation for further theoretical exploration.