arXiv:2601.21010v1 Announce Type: new Abstract: In this paper, we consider power consumption reduction in extremely large antenna arrays (ELAAs) for integrated sensing and communication (ISAC) applications. Although ELAAs are critical for achieving high-resolution near-field sensing, fully activating all antenna elements in conventional digital architectures leads to prohibitive power demands. To address this, we propose an energy-efficient subarray activation framework that selects an optimal subset of subarrays to minimize the total power consumption, subject to quality-of-service (QoS) constraints for […]

Ver mais

Reward design is of great importance for solving complex tasks with reinforcement learning. Recent studies have explored using image-text similarity produced by vision-language models (VLMs) to augment rewards of a task with visual feedback. A common practice linearly adds VLM scores to task or success rewards without explicit shaping, potentially altering the optimal policy. Moreover, such approaches, often relying on single static images, struggle with tasks whose desired behavior involves complex, dynamic motions spanning multiple visually different states. […]

Ver mais

Recently, TabPFN has gained attention as a foundation model for tabular data. However, it struggles to integrate heterogeneous modalities such as images and text, which are common in domains like healthcare and marketing, thereby limiting its applicability. To address this, we present the Multi-Modal Prior-data Fitted Network (MMPFN), which extends TabPFN to handle tabular and non-tabular modalities in a unified manner. MMPFN comprises per-modality encoders, modality projectors, and pre-trained foundation models. The modality projectors serve as the critical […]

Ver mais

arXiv:2602.04019v1 Announce Type: new Abstract: As large language models (LLMs) continue to grow, the cost of full-parameter fine-tuning has made parameter-efficient fine-tuning (PEFT) the default strategy for downstream adaptation. Constraints from inference latency in scalable serving and fine-tuning cost in edge or rapid-deployment settings make the choice of which layers to fine-tune unavoidable. Yet current practice typically applies PEFT uniformly across all layers, with limited understanding or leverage of layer selection. This paper develops a unified projected residual […]

Ver mais

arXiv:2601.21470v1 Announce Type: cross Abstract: We study semi-supervised stochastic optimization when labeled data is scarce but predictions from pre-trained models are available. PPI and SVRG both reduce variance through control variates — PPI uses predictions, SVRG uses reference gradients. We show they are mathematically equivalent and develop PPI-SVRG, which combines both. Our convergence bound decomposes into the standard SVRG rate plus an error floor from prediction uncertainty. The rate depends only on loss geometry; predictions affect only the […]

Ver mais

arXiv:2601.16382v1 Announce Type: new Abstract: While the filtered-x normalized least mean square (FxNLMS) algorithm is widely applied due to its simple structure and easy implementation for active noise control system, it faces two critical limitations: the fixed step-size causes a trade-off between convergence rate and steady-state residual error, and its performance deteriorates significantly in impulsive noise environments. To address the step-size constraint issue, we propose the switched mbox{step-size} FxNLMS (SSS-FxNLMS) algorithm. Specifically, we derive the mbox{mean-square} deviation (MSD) […]

Ver mais

Multi-stage ML inference pipelines are difficult to autoscale due to heterogeneous resources, cross-stage coupling, and dynamic bottleneck migration. We present SAIR, an autoscaling framework that uses an LLM as an in-context reinforcement learning controller, improving its policy online from reward-labeled interaction histories without gradient updates. SAIR combines Pareto-dominance reward shaping with a provable separation margin, surprisal-guided experience retrieval for context efficiency, and fine-grained GPU rate control via user-space CUDA interception. We provide regret analysis decomposing error into retrieval […]

Ver mais

More than 79,600 criminal cases were stuck in the court backlog in England and Wales at the end of 2025. The Crown Court backlog, in fact, has remained at record levels since early 2023 and, according to the Ministry of Justice, could reach 100,000 cases by 2028. Faced with overwhelmed courts and limited resources, governments are increasingly turning to artificial intelligence as a possible solution. AI is no longer a future concept for justice systems; it is already being used […]

Ver mais

arXiv:2602.09039v1 Announce Type: new Abstract: Prescriptive process monitoring seeks to recommend actions that improve process outcomes by analyzing possible continuations of ongoing cases. A key obstacle is the heavy computational cost of large-scale suffix comparisons, which grows rapidly with log size. We propose an efficient retrieval method exploiting the triangle inequality: distances to a set of optimized pivots define bounds that prune redundant comparisons. This substantially reduces runtime and is fully parallelizable. Crucially, pruning is exact: the retrieved […]

Ver mais

If there’s one database that promises both performance and cost and also delivers, then it’s Amazon DynamoDB. Single digit millisecond latency, fully managed with automatic scaling, pay per use – DynamoDB is genuinely impressive. But here’s the thing, there are rarely any mentions about what happens when your data model doesn’t fit DynamoDB’s worldview, or when you discover a limitation (I would rather say understand DynamoDB architecture better) three months into production that forces a complete redesign. I’ve […]

Ver mais