Diffusion Transformers (DiTs) have achieved state-of-the-art performance in generative modeling, yet their high computational cost hinders real-time deployment. While feature caching offers a promising training-free acceleration solution by exploiting temporal redundancy, existing methods suffer from two key limitations: (1) uniform caching intervals fail to align with the non-uniform temporal dynamics of DiT, and (2) naive feature reuse with excessively large caching intervals can lead to severe error accumulation. In this work, we analyze the evolution of DiT features during denoising and reveal that both feature changes and error propagation are highly time- and depth-varying. Motivated by this, we propose ProCache, a training-free dynamic feature caching framework that addresses these issues via two core components: (i) a constraint-aware caching pattern search module that generates non-uniform activation schedules through offline constrained sampling, tailored to the model’s temporal characteristics; and (ii) a selective computation module that selectively compute within deep blocks and high-importance tokens for cached segments to mitigate error accumulation with minimal overhead. Extensive experiments on PixArt-alpha and DiT demonstrate that ProCache achieves up to 1.96 times and 2.90 times acceleration with negligible quality degradation, significantly outperforming prior caching-based methods.

Modern information retrieval (IR) is no longer consumed primarily by humans but increasingly by large language models (LLMs) via retrieval-augmented generation (RAG) and agentic search. Unlike human users, LLMs are constrained by limited attention budgets and are uniquely vulnerable to noise; misleading or irrelevant information is no longer just a nuisance, but a direct cause of hallucinations and reasoning failures. In this perspective paper, we argue that denoising-maximizing usable evidence density and verifiability within a context window-is becoming the primary bottleneck across the full information access pipeline. We conceptualize this paradigm shift through a four-stage framework of IR challenges: from inaccessible to undiscoverable, to misaligned, and finally to unverifiable. Furthermore, we provide a pipeline-organized taxonomy of signal-to-noise optimization techniques, spanning indexing, retrieval, context engineering, verification, and agentic workflow. We also present research works on information denoising in domains that rely heavily on retrieval such as lifelong assistant, coding agent, deep research, and multimodal understanding.

Nowadays, data centers contribute to a large portion of global carbon emissions due to the growing demand for cloud services. Traditional green data center technologies have utilized all kinds of methods to reduce energy consumption, e.g., optimizing HVAC (Heating, Ventilation, and Air Conditioning), shutting down idle servers or switches and using renewable energy, but they typically target a single data center and have pushed the optimization space to its limits. However, carbon emissions vary wildly across data centers due to differences in local energy sources, and delay-tolerant tasks can be flexibly allocated to data centers with lower emissions. This opens new opportunities for carbon-aware task scheduling across geographical locations. However, carbon-aware scheduling faces key challenges, including the fluctuation of carbon intensity due to weather and power dynamics, and the dynamic nature of task arrivals with diverse QoS (Quality of Service) requirements. To address this, we propose CEDTS-RL (Carbon-Emission Driven Task Scheduling based on Reinforcement Learning), a real-time cross-geographical scheduling framework based on reinforcement learning, which unlocks new potential for reducing carbon emissions by relaxing the location restriction during scheduling. The evaluation is conducted using real-world datasets from Google, Alibaba, and Microsoft, along with solar and meteorological data from NASA. Results demonstrate that CEDTS-RL significantly outperforms both traditional and recent scheduling algorithms.