We present LazyDINO, a transport map variational inference method for fast, scalable, and efficiently amortized solutions of high-dimensional nonlinear Bayesian inverse problems with expensive parameter-to-observable (PtO) maps. Our method consists of an offline phase in which we construct a derivative-informed neural surrogate of the PtO map using joint samples of the PtO map and its Jacobian. During the online phase, when given observational data, we seek rapid posterior approximation using surrogate-driven training of a lazy map [Brennan et al., NeurIPS, (2020)], i.e., a structure-exploiting transport map with low-dimensional nonlinearity. The trained lazy map then produces approximate posterior samples or density evaluations. Our surrogate construction is optimized for amortized Bayesian inversion using lazy map variational inference. We show that (i) the derivative-based reduced basis architecture [O'Leary-Roseberry et al., Comput. Methods Appl. Mech. Eng., 388 (2022)] minimizes the upper bound on the expected error in surrogate posterior approximation, and (ii) the derivative-informed training formulation [O'Leary-Roseberry et al., J. Comput. Phys., 496 (2024)] minimizes the expected error due to surrogate-driven transport map optimization. Our numerical results demonstrate that LazyDINO is highly efficient in cost amortization for Bayesian inversion. We observe one to two orders of magnitude reduction of offline cost for accurate posterior approximation, compared to simulation-based amortized inference via conditional transport and conventional surrogate-driven transport. In particular, LazyDINO outperforms Laplace approximation consistently using fewer than 1000 offline samples, while other amortized inference methods struggle and sometimes fail at 16,000 offline samples.

The predict-then-optimize (PTO) framework is indispensable for addressing practical stochastic decision-making tasks. It consists of two crucial steps: initially predicting unknown parameters of an optimization model and subsequently solving the problem based on these predictions. Elmachtoub and Grigas [1] introduced the Smart Predict-then-Optimize (SPO) loss for the framework, which gauges the decision error arising from predicted parameters, and a convex surrogate, the SPO+ loss, which incorporates the underlying structure of the optimization model. The consistency of these different loss functions is guaranteed under the assumption of i.i.d. training data. Nevertheless, various types of data are often dependent, such as power load fluctuations over time. This dependent nature can lead to diminished model performance in testing or real-world applications. Motivated to make intelligent predictions for time series data, we present an autoregressive SPO method directly targeting the optimization problem at the decision stage in this paper, where the conditions of consistency are no longer met. Therefore, we first analyze the generalization bounds of the SPO loss within our autoregressive model. Subsequently, the uniform calibration results in Liu and Grigas [2] are extended in the proposed model. Finally, we conduct experiments to empirically demonstrate the effectiveness of the SPO+ surrogate compared to the absolute loss and the least squares loss, especially when the cost vectors are determined by stationary dynamical systems and demonstrate the relationship between normalized regret and mixing coefficients.

By the end of 2023, renewable sources cover 63.4% of the total electric power demand of Chile, and in line with the global trend, photovoltaic (PV) power shows the most dynamic increase. Although Chile's Atacama Desert is considered the sunniest place on Earth, PV power production, even in this area, can be highly volatile. Successful integration of PV energy into the country's power grid requires accurate short-term PV power forecasts, which can be obtained from predictions of solar irradiance and related weather quantities. Nowadays, in weather forecasting, the state-of-the-art approach is the use of ensemble forecasts based on multiple runs of numerical weather prediction models. However, ensemble forecasts still tend to be uncalibrated or biased, thus requiring some form of post-processing. The present work investigates probabilistic forecasts of solar irradiance for Regions III and IV in Chile. For this reason, 8-member short-term ensemble forecasts of solar irradiance for calendar year 2021 are generated using the Weather Research and Forecasting (WRF) model, which are then calibrated using the benchmark ensemble model output statistics (EMOS) method based on a censored Gaussian law, and its machine learning-based distributional regression network (DRN) counterpart. Furthermore, we also propose a neural network-based post-processing method resulting in improved 8-member ensemble predictions. All forecasts are evaluated against station observations for 30 locations, and the skill of post-processed predictions is compared to the raw WRF ensemble. Our case study confirms that all studied post-processing methods substantially improve both the calibration of probabilistic- and the accuracy of point forecasts. Among the methods tested, the corrected ensemble exhibits the best overall performance. Additionally, the DRN model generally outperforms the corresponding EMOS approach.

Despite the remarkable capabilities of modern large language models (LLMs), the mechanisms behind their problem-solving abilities remain elusive. In this work, we aim to better understand how the learning dynamics of LLM finetuning shapes downstream generalization. Our analysis focuses on reasoning tasks, whose problem structure allows us to distinguish between memorization (the exact replication of reasoning steps from the training data) and performance (the correctness of the final solution). We find that a model's generalization behavior can be effectively characterized by a training metric we call pre-memorization train accuracy: the accuracy of model samples on training queries before they begin to copy the exact reasoning steps from the training set. On the dataset level, this metric is able to reliably predict test accuracy, achieving $R^2$ of around or exceeding 0.9 across various models (Llama3 8, Gemma2 9B), datasets (GSM8k, MATH), and training configurations. On a per-example level, this metric is also indicative of whether individual model predictions are robust to perturbations in the training query. By connecting a model's learning behavior to its generalization, pre-memorization train accuracy can guide targeted improvements to training strategies. We focus on data curation as an example, and show that prioritizing examples with low pre-memorization accuracy leads to 1.5-2x improvements in data efficiency compared to i.i.d. data scaling, and outperforms other standard data curation techniques.

Accurate urban microclimate analysis with wind velocity and temperature is vital for energy-efficient urban planning, supporting carbon reduction, enhancing public health and comfort, and advancing the low-altitude economy. However, traditional computational fluid dynamics (CFD) simulations that couple velocity and temperature are computationally expensive. Recent machine learning advancements offer promising alternatives for accelerating urban microclimate simulations. The Fourier neural operator (FNO) has shown efficiency and accuracy in predicting single-variable velocity magnitudes in urban wind fields. Yet, for multivariable high-resolution 3D urban microclimate prediction, FNO faces three key limitations: blurry output quality, high GPU memory demand, and substantial data requirements. To address these issues, we propose a novel localized Fourier neural operator (Local-FNO) model that employs local training, geometry encoding, and patch overlapping. Local-FNO provides accurate predictions for rapidly changing turbulence in urban microclimate over 60 seconds, four times the average turbulence integral time scale, with an average error of 0.35 m/s in velocity and 0.30 {\deg}C in temperature. It also accurately captures turbulent heat flux represented by the velocity-temperature correlation. In a 2 km by 2 km domain, Local-FNO resolves turbulence patterns down to a 10 m resolution. It provides high-resolution predictions with 150 million feature dimensions on a single 32 GB GPU at nearly 50 times the speed of a CFD solver. Compared to FNO, Local-FNO achieves a 23.9% reduction in prediction error and a 47.3% improvement in turbulent fluctuation correlation.

Deep learning models have shown strong performance in load forecasting, but they generally require large amounts of data for model training before being applied to new scenarios, which limits their effectiveness in data-scarce scenarios. Inspired by the great success of pre-trained language models (LLMs) in natural language processing, this paper proposes a zero-shot load forecasting approach using an advanced LLM framework denoted as the Chronos model. By utilizing its extensive pre-trained knowledge, the Chronos model enables accurate load forecasting in data-scarce scenarios without the need for extensive data-specific training. Simulation results across five real-world datasets demonstrate that the Chronos model significantly outperforms nine popular baseline models for both deterministic and probabilistic load forecasting with various forecast horizons (e.g., 1 to 48 hours), even though the Chronos model is neither tailored nor fine-tuned to these specific load datasets. Notably, Chronos reduces root mean squared error (RMSE), continuous ranked probability score (CRPS), and quantile score (QS) by approximately 7.34%-84.30%, 19.63%-60.06%, and 22.83%-54.49%, respectively, compared to baseline models. These results highlight the superiority and flexibility of the Chronos model, positioning it as an effective solution in data-scarce scenarios.

Quadrotors equipped with cable-suspended loads represent a versatile, low-cost, and energy efficient solution for aerial transportation, construction, and manipulation tasks. However, their real-world deployment is hindered by several challenges. The system is difficult to control because it is nonlinear, underactuated, involves hybrid dynamics due to slack-taut cable modes, and evolves on complex configuration spaces. Additionally, it is crucial to estimate the full state and the cable's mode transitions in real-time using on-board sensors and computation. To address these challenges, we present a novel Hybrid Perception-Aware Nonlinear Model Predictive Control (HPA-MPC) control approach for quadrotors with suspended loads. Our method considers the complete hybrid system dynamics and includes a perception-aware cost to ensure the payload remains visible in the robot's camera during navigation. Furthermore, the full state and hybrid dynamics' transitions are estimated using onboard sensors. Experimental results demonstrate that our approach enables stable load tracking control, even during slack-taut transitions, and operates entirely onboard. The experiments also show that the perception-aware term effectively keeps the payload in the robot's camera field of view when a human operator interacts with the load.

Traditional Learning-To-Rank (LETOR) approaches, including pairwise methods like RankNet and LambdaMART, often fall short by solely focusing on pairwise comparisons, leading to sub-optimal global rankings. Conversely, deep learning based listwise methods, while aiming to optimise entire lists, require complex tuning and yield only marginal improvements over robust pairwise models. To overcome these limitations, we introduce Travelling Salesman Problem Rank (TSPRank), a hybrid pairwise-listwise ranking method. TSPRank reframes the ranking problem as a Travelling Salesman Problem (TSP), a well-known combinatorial optimisation challenge that has been extensively studied for its numerous solution algorithms and applications. This approach enables the modelling of pairwise relationships and leverages combinatorial optimisation to determine the listwise ranking. This approach can be directly integrated as an additional component into embeddings generated by existing backbone models to enhance ranking performance. Our extensive experiments across three backbone models on diverse tasks, including stock ranking, information retrieval, and historical events ordering, demonstrate that TSPRank significantly outperforms both pure pairwise and listwise methods. Our qualitative analysis reveals that TSPRank's main advantage over existing methods is its ability to harness global information better while ranking. TSPRank's robustness and superior performance across different domains highlight its potential as a versatile and effective LETOR solution. The code and preprocessed data are available at https://github.com/waylonli/TSPRank-KDD2025.

Fragment-based drug discovery, in which molecular fragments are assembled into new molecules with desirable biochemical properties, has achieved great success. However, many fragment-based molecule generation methods show limited exploration beyond the existing fragments in the database as they only reassemble or slightly modify the given ones. To tackle this problem, we propose a new fragment-based molecule generation framework with retrieval augmentation, namely Fragment Retrieval-Augmented Generation (f-RAG). f-RAG is based on a pre-trained molecular generative model that proposes additional fragments from input fragments to complete and generate a new molecule. Given a fragment vocabulary, f-RAG retrieves two types of fragments: (1) hard fragments, which serve as building blocks that will be explicitly included in the newly generated molecule, and (2) soft fragments, which serve as reference to guide the generation of new fragments through a trainable fragment injection module. To extrapolate beyond the existing fragments, f-RAG updates the fragment vocabulary with generated fragments via an iterative refinement process which is further enhanced with post-hoc genetic fragment modification. f-RAG can achieve an improved exploration-exploitation trade-off by maintaining a pool of fragments and expanding it with novel and high-quality fragments through a strong generative prior.

Camera traps, combined with AI, have emerged as a way to achieve automated, scalable biodiversity monitoring. However, the passive infrared (PIR) sensors that trigger camera traps are poorly suited for detecting small, fast-moving ectotherms such as insects. Insects comprise over half of all animal species and are key components of ecosystems and agriculture. The need for an appropriate and scalable insect camera trap is critical in the wake of concerning reports of declines in insect populations. This study proposes an alternative to the PIR trigger: ultra-lightweight convolutional neural networks running on low-powered hardware to detect insects in a continuous stream of captured images. We train a suite of models to distinguish insect images from backgrounds. Our design achieves zero latency between trigger and image capture. Our models are rigorously tested and achieve high accuracy ranging from 91.8% to 96.4% AUC on validation data and >87% AUC on data from distributions unseen during training. The high specificity of our models ensures minimal saving of false positive images, maximising deployment storage efficiency. High recall scores indicate a minimal false negative rate, maximising insect detection. Further analysis with saliency maps shows the learned representation of our models to be robust, with low reliance on spurious background features. Our system is also shown to operate deployed on off-the-shelf, low-powered microcontroller units, consuming a maximum power draw of less than 300mW. This enables longer deployment times using cheap and readily available battery components. Overall we offer a step change in the cost, efficiency and scope of insect monitoring. Solving the challenging trigger problem, we demonstrate a system which can be deployed for far longer than existing designs and budgets power and bandwidth effectively, moving towards a generic insect camera trap.