The rapid advancements in satellite remote sensing have enhanced the capability to monitor and analyze the Earth's surface. Among the many variables captured through satellite sensors, Land Surface Temperature (LST) plays a critical role in understanding key environmental processes. However, obtaining high-resolution LST data remains a challenge, as satellite sensors often face a trade-off between spatial and temporal resolutions. In response, Spatio-Temporal Fusion (STF) has emerged as a powerful method to integrate two satellite data sources, one providing high spatial but low temporal resolution, and the other offering high temporal but low spatial resolution. Although a range of STF techniques have been proposed, from traditional methods to cutting-edge deep learning (DL) models, most have focused on surface reflectance, with limited application to LST estimation. DL approaches, in particular, show promise in improving the spatial and temporal resolutions of LST by capturing complex, non-linear relationships between input and output LST data. This paper offers a comprehensive review of the latest advancements in DL-based STF techniques for LST estimation. We analyze key research developments, mathematically formulate the STF problem, and introduce a novel taxonomy for DL-based STF methods. Furthermore, we discuss the challenges faced by current methods and highlight future research directions. In addition, we present the first open-source benchmark STF dataset for LST estimation, consisting of 51 pairs of MODIS-Landsat images spanning from 2013 to 2024. To support our findings, we conduct extensive experiments on state-of-the-art methods and present both quantitative and qualitative assessments. This is the first survey paper focused on DL-based STF for LST estimation. We hope it serves as a valuable reference for researchers and paves the way for future research in this field.

Long-term time-series forecasting is essential for planning and decision-making in economics, energy, and transportation, where long foresight is required. To obtain such long foresight, models must be both efficient and effective in processing long sequence. Recent advancements have enhanced the efficiency of these models; however, the challenge of effectively leveraging longer sequences persists. This is primarily due to the tendency of these models to overfit when presented with extended inputs, necessitating the use of shorter input lengths to maintain tolerable error margins. In this work, we investigate the multiscale modeling method and propose the Logsparse Decomposable Multiscaling (LDM) framework for the efficient and effective processing of long sequences. We demonstrate that by decoupling patterns at different scales in time series, we can enhance predictability by reducing non-stationarity, improve efficiency through a compact long input representation, and simplify the architecture by providing clear task assignments. Experimental results demonstrate that LDM not only outperforms all baselines in long-term forecasting benchmarks, but also reducing both training time and memory costs.

Nonetheless, the technological advancement of AI relies on computationally intensive calculations and thus has led to a surge in resource usage and energy consumption. Even putting aside the environmental toll of server manufacturing and supply chains, AI systems can create a huge environmental cost to communities and regions where they are deployed, including air/thermal pollution due to fossil fuel-based electricity generation and further stressed water resources due to AI's staggering water footprint [12, 25]. To make AI more environmentally friendly and ensure that its overall impacts on climate change are positive, recent studies have pursued multi-faceted approaches, including efficient training and inference [5], energy-efficient GPU and accelerator designs [19], carbon forecasting[14], carbon-aware task scheduling[1, 21], green cloud infrastructures[2], sustainable AI policies [10, 18], and more. Additionally, data center operators have also increasingly adopted carbon-free energy(such as solar and wind power) and climate-conscious cooling systems, lowering carbon footprint and direct water consumption [8]. Although these initiatives are encouraging, unfortunately, a worrisome outcome-- environmental inequity -- has emerged [3]. That is, minimizing the total environmental cost of a globally deployed AI system across multiple regions does not necessarily mean that each region is treated equitably. In fact, the environmental cost of AI is often disproportionately higher in certain disadvantaged regions than in others. Even worse, AI's environmental inequity can be amplified by existing environmental equity agnostic resource allocation, load balancing, and scheduling algorithms and compounded by enduring socioeconomic disparities between regions.

Inspired by the Kolmogorov-Arnold representation theorem and Kurkova's principle of using approximate representations, we propose the Kurkova-Kolmogorov-Arnold Network (KKAN), a new two-block architecture that combines robust multi-layer perceptron (MLP) based inner functions with flexible linear combinations of basis functions as outer functions. We first prove that KKAN is a universal approximator, and then we demonstrate its versatility across scientific machine-learning applications, including function regression, physics-informed machine learning (PIML), and operator-learning frameworks. The benchmark results show that KKANs outperform MLPs and the original Kolmogorov-Arnold Networks (KANs) in function approximation and operator learning tasks and achieve performance comparable to fully optimized MLPs for PIML. To better understand the behavior of the new representation models, we analyze their geometric complexity and learning dynamics using information bottleneck theory, identifying three universal learning stages, fitting, transition, and diffusion, across all types of architectures. We find a strong correlation between geometric complexity and signal-to-noise ratio (SNR), with optimal generalization achieved during the diffusion stage. Additionally, we propose self-scaled residual-based attention weights to maintain high SNR dynamically, ensuring uniform convergence and prolonged learning.

We introduce MLE-bench, a benchmark for measuring how well AI agents perform at machine learning engineering. To this end, we curate 75 ML engineering-related competitions from Kaggle, creating a diverse set of challenging tasks that test real-world ML engineering skills such as training models, preparing datasets, and running experiments. We establish human baselines for each competition using Kaggle's publicly available leaderboards. We use open-source agent scaffolds to evaluate several frontier language models on our benchmark, finding that the best-performing setup--OpenAI's o1-preview with AIDE scaffolding--achieves at least the level of a Kaggle bronze medal in 16.9% of competitions. In addition to our main results, we investigate various forms of resource scaling for AI agents and the impact of contamination from pre-training. We open-source our benchmark code (github.com/openai/mle-bench/) to facilitate future research in understanding the ML engineering capabilities of AI agents.

Enterprises possess a vast array of API assets scattered across various functions, forming the backbone of existing business processes. By leveraging these APIs as functional tools, enterprises can design diverse, scenario-specific agent applications, driven by on-premise function-calling models as the core engine. However, generic models often fail to meet enterprise requirements in terms of computational efficiency, output accuracy, and stability, necessitating scenario-specific adaptation. In this paper, we propose a training pipeline for function-calling capabilities tailored to real-world business scenarios. This pipeline includes the synthesis and augmentation of scenario-specific function-calling data, model fine-tuning, and performance evaluation and analysis. Using this pipeline, we generated 1,260 fully AI-generated samples and 1,035 augmented manually-labeled samples in digital HR agent scenario. The Qwen2.5-Coder-7B-Instruct model was employed as the base model and fine-tuned using the LoRA method on four GPUs with 24GB VRAM. Our fine-tuned model demonstrated outstanding performance in evaluations and practical applications, surpassing GPT-4 and GPT-4o in accuracy on the test set. These results validate the reliability of the proposed pipeline for training scenario-specific function-calling models.

Bayesian Optimization (BO) is a widely-used method for optimizing expensive-to-evaluate black-box functions. Traditional BO assumes that the learner has full control over all query variables without additional constraints. However, in many real-world scenarios, controlling certain query variables may incur costs. Therefore, the learner needs to balance the selection of informative subsets for targeted learning against leaving some variables to be randomly sampled to minimize costs. This problem is known as Bayesian Optimization with cost-varying variable subsets (BOCVS). While the goal of BOCVS is to identify the optimal solution with minimal cost, previous works have only guaranteed finding the optimal solution without considering the total costs incurred. Moreover, these works assume precise knowledge of the cost for each subset, which is often unrealistic. In this paper, we propose a novel algorithm for the extension of the BOCVS problem with random and unknown costs that separates the process into exploration and exploitation phases. The exploration phase will filter out low-quality variable subsets, while the exploitation phase will leverage high-quality ones. Furthermore, we theoretically demonstrate that our algorithm achieves a sub-linear rate in both quality regret and cost regret, addressing the objective of the BOCVS problem more effectively than previous analyses. Finally, we show that our proposed algorithm outperforms comparable baselines across a wide range of benchmarks.

The discovery of novel mechanical metamaterials, whose properties are dominated by their engineered structures rather than chemical composition, is a knowledge-intensive and resource-demanding process. To accelerate the design of novel metamaterials, we present MetaScientist, a human-in-the-loop system that integrates advanced AI capabilities with expert oversight with two primary phases: (1) hypothesis generation, where the system performs complex reasoning to generate novel and scientifically sound hypotheses, supported with domain-specific foundation models and inductive biases retrieved from existing literature; (2) 3D structure synthesis, where a 3D structure is synthesized with a novel 3D diffusion model based on the textual hypothesis and refined it with a LLM-based refinement model to achieve better structure properties. At each phase, domain experts iteratively validate the system outputs, and provide feedback and supplementary materials to ensure the alignment of the outputs with scientific principles and human preferences. Through extensive evaluation from human scientists, MetaScientist is able to deliver novel and valid mechanical metamaterial designs that have the potential to be highly impactful in the metamaterial field.

We propose a new simulator, training approach, and policy architecture, collectively called SOUS VIDE, for end-to-end visual drone navigation. Our trained policies exhibit zero-shot sim-to-real transfer with robust real-world performance using only on-board perception and computation. Our simulator, called FiGS, couples a computationally simple drone dynamics model with a high visual fidelity Gaussian Splatting scene reconstruction. FiGS can quickly simulate drone flights producing photorealistic images at up to 130 fps. We use FiGS to collect 100k-300k observation-action pairs from an expert MPC with privileged state and dynamics information, randomized over dynamics parameters and spatial disturbances. We then distill this expert MPC into an end-to-end visuomotor policy with a lightweight neural architecture, called SV-Net. SV-Net processes color image, optical flow and IMU data streams into low-level body rate and thrust commands at 20Hz onboard a drone. Crucially, SV-Net includes a Rapid Motor Adaptation (RMA) module that adapts at runtime to variations in drone dynamics. In a campaign of 105 hardware experiments, we show SOUS VIDE policies to be robust to 30% mass variations, 40 m/s wind gusts, 60% changes in ambient brightness, shifting or removing objects from the scene, and people moving aggressively through the drone's visual field. Code, data, and experiment videos can be found on our project page: https://stanfordmsl.github.io/SousVide/.

Fluid flow is a widely applied physical problem, crucial in various fields. Due to the highly nonlinear and chaotic nature of fluids, analyzing fluid-related problems is exceptionally challenging. Computational fluid dynamics (CFD) is the best tool for this analysis but involves significant computational resources, especially for 3D simulations, which are slow and resource-intensive. In experimental fluid dynamics, PIV cost increases with dimensionality. Reconstructing 3D flow fields from 2D PIV data could reduce costs and expand application scenarios. Here, We propose a Diffusion Transformer-based method for reconstructing 3D flow fields from 2D flow data. By embedding the positional information of 2D planes into the model, we enable the reconstruction of 3D flow fields from any combination of 2D slices, enhancing flexibility. We replace global attention with window and plane attention to reduce computational costs associated with higher dimensions without compromising performance. Our experiments demonstrate that our model can efficiently and accurately reconstruct 3D flow fields from 2D data, producing realistic results.