arXiv subjects

Haochen Wu

Publications and source records attributed to Haochen Wu.

At least 19 recordsLinked to original sources

arXiv2026

Multi-Agent Reinforcement Learning from Delayed Marketplace Feedback for Objective-Weight Adaptation in Three-Sided Dispatch

Dispatch in three-sided marketplaces provides a natural setting for reinforcement learning from world feedback: decisions are evaluated by delayed operational outcomes such as delivery speed, courier utilization, and merchant congestion. We present a deployed reinforcement learning system at DoorDash that adapts dispatch objective weights in a large-scale food-delivery marketplace using delayed signals. Rather than replacing the combinatorial assignment optimizer, a store-level policy learned from logged marketplace data selects a discrete multiplier that shifts the dispatch optimizer's tradeoff between delivery quality and batching efficiency. This interface enables offline policy learning under noisy, delayed, and coupled feedback while preserving production feasibility constraints and operational safeguards. We train a shared value function using centralized offline data and decentralized store-level execution, with Double Q-learning targets and a conservative regularizer to reduce out-of-distribution value overestimation. In a production switchback experiment, the offline-trained policy increases batching and reduces courier-side time costs without degrading customer-facing delivery quality. Results illustrate how world feedback from a live economic and logistics system can be used to safely adapt decision policies online.

cs.AI↗

arXiv2026

Operator learning on domain boundary through combining fundamental solution-based artificial data and boundary integral techniques

For linear partial differential equations with known fundamental solutions, this work introduces a novel operator learning framework that relies exclusively on domain boundary data, including solution values and normal derivatives, rather than full-domain sampling. By integrating the previously developed Mathematical Artificial Data (MAD) method, which enforces physical consistency, all training data are synthesized directly from the fundamental solutions of the target problems, resulting in a fully data-driven pipeline without the need for external measurements or numerical simulations. We refer to this approach as the Mathematical Artificial Data Boundary Neural Operator (MAD-BNO), which learns boundary-to-boundary mappings using MAD-generated Dirichlet-Neumann data pairs. Once trained, the interior solution at arbitrary locations can be efficiently recovered through boundary integral formulations, supporting Dirichlet, Neumann, and mixed boundary conditions as well as general source terms. The proposed method is validated on benchmark operator learning tasks for two-dimensional Laplace, Poisson, and Helmholtz equations, where it achieves accuracy comparable to or better than existing neural operator approaches while significantly reducing training time. The framework is naturally extensible to three-dimensional problems and complex geometries.

cs.LG↗

arXiv2025

Hierarchically Block-Sparse Recovery With Prior Support Information

We provide new recovery bounds for hierarchical compressed sensing (HCS) based on prior support information (PSI). A detailed PSI-enabled reconstruction model is formulated using various forms of PSI. The hierarchical block orthogonal matching pursuit with PSI (HiBOMP-P) algorithm is designed in a recursive form to reliably recover hierarchically block-sparse signals. We derive exact recovery conditions (ERCs) measured by the mutual incoherence property (MIP), wherein hierarchical MIP concepts are proposed, and further develop reconstructible sparsity levels to reveal sufficient conditions for ERCs. Leveraging these MIP analyses, we present several extended insights, including reliable recovery conditions in noisy scenarios and the optimal hierarchical structure for cases where sparsity is not equal to zero. Our results further confirm that HCS offers improved recovery performance even when the prior information does not overlap with the true support set, whereas existing methods heavily rely on this overlap, thereby compromising performance if it is absent.

eess.SP↗

arXiv2025

Two-timescale Beamforming Optimization for Downlink Multi-user Holographic MIMO Surfaces

Benefiting from the rapid development of metamaterials and metasurfaces, the holographic multiple-input and multiple-output surface (HMIMOS) has been regarded as a promising solution for future wireless networks recently. By densely packing numerous radiation elements together, HMIMOS is capable of realizing accurate beamforming with low hardware complexity. However, enormous radiation elements on the HMIMOS lead to high computational complexity and signaling overhead when applying traditional beamforming schemes relying on instantaneous channel state information (CSI). To tackle this problem, we propose a two-timescale optimization scheme to minimize the required transmission power under the constraint of all users' quality-of-service (QoS). Specifically, the beampatterns at the base station (BS) and the user equippment (UE) are optimized over the slowly changing statistical CSI based on the constrained stochastic successive convex approximation (CSSCA) algorithm. Then, the instantaneous CSI is utilized to design the precoding matrix in order to ensure the system performance without significant increase in computational cost, due to the small number of feeds on the HMIMOS. Simulation results demonstrate the effectiveness of our proposed method compared to other baselines.

eess.SP↗

arXiv2025

An orthogonal perspective on Gauss composition

We revisit Gauss composition over a general base scheme, with a focus on orthogonal groups. We show that the Clifford and norm functors provide an equivalence of stacks between binary quadratic modules and pseudoregular modules over quadratic algebras. As a consequence, we exhibit a composition law for coprimitive forms over a general base, including a universal version of Dirichlet composition. This perspective synthesizes the constructions of Kneser and Wood, reconciles algebraic and geometric approaches, and clarifies the role of orientations and the natural emergence of narrow class groups.

math.RA↗

arXiv2025

Integrated Prediction and Distributionally Robust Optimization for Air Traffic Management

Strategic Traffic Management Initiatives (TMIs) such as Ground Delay Programs (GDPs) play a crucial role in mitigating operational costs associated with air traffic demand-capacity imbalances. However, GDPs can only be planned (e.g., duration, delay assignments) with confidence if the future capacities at constrained resources (i.e., airports) are predictable. In reality, such future capacities are uncertain, and predictive models may provide predictions that are vulnerable to errors and distribution shifts. Motivated by the goal of planning optimal GDPs that are distributionally robust against airport capacity prediction errors, we study a fully integrated learning-driven optimization framework. We design a deep learning-based prediction model capable of forecasting arrival and departure capacity distributions across a network of airports. We incorporate the predictions into a distributionally robust formulation of the multi-airport ground holding program (DR-MAGHP). Our results demonstrate that DR-MAGHP can achieve up to a 15.6% improvement over the stochastic programming formulation (SP-MAGHP) under airport capacity distribution shifts. We conclude by outlining future research directions aimed at enhancing both the learning and optimization components of the framework.

math.OC↗

arXiv2025

Optimization-Guided Exploration of Advanced Air Mobility Congestion Management Strategies with Stochastic Demands

Advanced Air Mobility (AAM) represents an evolution of the air transportation system by introducing low-altitude, potentially high-traffic environments. AAM operations will be enabled by both new aircraft, as well as new safety- and efficiency-critical supporting infrastructure. Published concepts of operations from both public and private sector entities establish notions such as federated management of the airspace and public-private partnerships for AAM air traffic, but there is a gap in the literature in terms of integrated tools that consider all three critical elements: AAM fleet operators (\emph{lower} layer), airspace service providers (\emph{middle} layer), and overall system governance from the legacy air navigation service provider (\emph{upper} layer). In this work, we explore modeling congestion management within the AAM setting using a bi-level optimization approach, focusing on (1) time-varying, stochastic AAM demand, (2) differing congestion management strategies, and (3) the impact of unscheduled, \enquote{pop-up} demand. We show that our bi-level formulation can be tractably solved using a Neural Network-based surrogate which returns solution qualities within 0.1-5.2\% of the optimal solution. Additionally, we show that our congestion management strategies can reduce congestion by 25.7-39.8\% when compared to the scenario of no strategies being applied. Finally, we also show that while pop-up demand degrades congestion conditions, our congestion management strategies fare better against pop-up demand than the no strategy scenario. The work herein contributes a rigorous modeling and simulation tool to help evaluate future AAM traffic management concepts and strategies.

math.OC↗

arXiv2025

New Paradigm for Unified Near-Field and Far-Field Wireless Communications

Current Type I and Type II codebooks in fifth generation (5G) wireless communications are limited in supporting the coexistence of far-field and near-field user equipments, as they are exclusively designed for far-field scenarios. To fill this knowledge gap and encourage relevant proposals by the 3rd Generation Partnership Project (3GPP), this article provides a novel codebook to facilitate a unified paradigm for the coexistence of far-field and near-field contexts. It ensures efficient precoding for all user equipments (UEs), while removing the need for the base station to identify whether one specific UE stays in either near-field or far-field regions. Additionally, our proposed codebook ensures compliance with current 3GPP standards for working flow and reference signals. Simulation results demonstrate the superior performance and versatility of our proposed codebook, validating its effectiveness in unifying near-field and far-field precoding for sixth-generation (6G) multiple-input multiple-output (MIMO) systems.

eess.SP↗

arXiv2024

ConDSeg: A General Medical Image Segmentation Framework via Contrast-Driven Feature Enhancement

Medical image segmentation plays an important role in clinical decision making, treatment planning, and disease tracking. However, it still faces two major challenges. On the one hand, there is often a ``soft boundary'' between foreground and background in medical images, with poor illumination and low contrast further reducing the distinguishability of foreground and background within the image. On the other hand, co-occurrence phenomena are widespread in medical images, and learning these features is misleading to the model's judgment. To address these challenges, we propose a general framework called Contrast-Driven Medical Image Segmentation (ConDSeg). First, we develop a contrastive training strategy called Consistency Reinforcement. It is designed to improve the encoder's robustness in various illumination and contrast scenarios, enabling the model to extract high-quality features even in adverse environments. Second, we introduce a Semantic Information Decoupling module, which is able to decouple features from the encoder into foreground, background, and uncertainty regions, gradually acquiring the ability to reduce uncertainty during training. The Contrast-Driven Feature Aggregation module then contrasts the foreground and background features to guide multi-level feature fusion and key feature enhancement, further distinguishing the entities to be segmented. We also propose a Size-Aware Decoder to solve the scale singularity of the decoder. It accurately locate entities of different sizes in the image, thus avoiding erroneous learning of co-occurrence features. Extensive experiments on five medical image datasets across three scenarios demonstrate the state-of-the-art performance of our method, proving its advanced nature and general applicability to various medical image segmentation scenarios. Our released code is available at \url{https://github.com/Mengqi-Lei/ConDSeg}.

eess.IV↗

arXiv2024

USTC-KXDIGIT System Description for ASVspoof5 Challenge

This paper describes the USTC-KXDIGIT system submitted to the ASVspoof5 Challenge for Track 1 (speech deepfake detection) and Track 2 (spoofing-robust automatic speaker verification, SASV). Track 1 showcases a diverse range of technical qualities from potential processing algorithms and includes both open and closed conditions. For these conditions, our system consists of a cascade of a frontend feature extractor and a back-end classifier. We focus on extensive embedding engineering and enhancing the generalization of the back-end classifier model. Specifically, the embedding engineering is based on hand-crafted features and speech representations from a self-supervised model, used for closed and open conditions, respectively. To detect spoof attacks under various adversarial conditions, we trained multiple systems on an augmented training set. Additionally, we used voice conversion technology to synthesize fake audio from genuine audio in the training set to enrich the synthesis algorithms. To leverage the complementary information learned by different model architectures, we employed activation ensemble and fused scores from different systems to obtain the final decision score for spoof detection. During the evaluation phase, the proposed methods achieved 0.3948 minDCF and 14.33% EER in the close condition, and 0.0750 minDCF and 2.59% EER in the open condition, demonstrating the robustness of our submitted systems under adversarial conditions. In Track 2, we continued using the CM system from Track 1 and fused it with a CNN-based ASV system. This approach achieved 0.2814 min-aDCF in the closed condition and 0.0756 min-aDCF in the open condition, showcasing superior performance in the SASV system.

cs.SD↗

arXiv2024

Near-Field Channel Estimation in Dual-Band XL-MIMO with Side Information-Assisted Compressed Sensing

Near-field communication comes to be an indispensable part of the future sixth generation (6G) communications at the arrival of the forth-coming deployment of extremely large-scale multiple-input-multiple-output (XL-MIMO) systems. Due to the huge array aperture and high-frequency bands, the electromagnetic radiation field is modeled by the spherical waves instead of the conventional planar waves, leading to severe weak sparsity to angular-domain near-field channel. Therefore, the channel estimation reminiscent of the conventional compression sensing (CS) approaches in the angular domain, judiciously utilized for low pilot overhead, may result in unprecedented challenges. To this end, this paper proposes a brand-new near-field channel estimation scheme by exploiting the naturally occurring useful side information. Specifically, we formulate the dual-band near-field communication model based on the fact that high-frequency systems are likely to be deployed with lower-frequency systems. Representative side information, i.e., the structural characteristic information derived by the sparsity ambiguity and the out-of-band spatial information stemming from the lower-frequency channel, is explored and tailored to materialize exceptional near-field channel estimation. Furthermore, in-depth theoretical analyses are developed to guarantee the minimum estimation error, based on which a suite of algorithms leveraging the elaborating side information are proposed. Numerical simulations demonstrate that the designed algorithms provide more assured results than the off-the-shelf approaches in the context of the dual-band near-field communications in both on- and off-grid scenarios, where the angle of departures/arrivals are discretely or continuously distributed, respectively.

eess.SP↗

arXiv2024

Distributionally Robust Ground Delay Programs with Learning-Driven Airport Capacity Predictions

Strategic Traffic Management Initiatives (TMIs) such as Ground Delay Programs (GDPs) play a crucial role in mitigating operational costs associated with demand-capacity imbalances. However, GDPs can only be planned (e.g., duration, delay assignments) with confidence if the future capacities at constrained resources (i.e., airports) are predictable. In reality, such future capacities are uncertain, and predictive models may provide forecasts that are vulnerable to errors and distribution shifts. Motivated by the goal of planning optimal GDPs that are \emph{distributionally robust} against airport capacity prediction errors, we study a fully integrated learning-driven optimization framework. We design a deep learning-based prediction model capable of forecasting arrival and departure capacity distributions across a network of airports. We then integrate the forecasts into a distributionally robust formulation of the multi-airport ground holding problem (\textsc{dr-MAGHP}). We show how \textsc{dr-MAGHP} can outperform stochastic optimization when distribution shifts occur, and conclude with future research directions to improve both the learning and optimization stages.

math.OC↗

arXiv2024

Spaceport Facility Location Planning within the US National Airspace System

The burgeoning commercial space transportation industry necessitates an expansion of launch infrastructure to meet rising demands. However, future operations from these large-scale infrastructures can result in new impacts, particularly to air traffic operations. To rigorously reason about where such future spaceports might be located and what their impacts might be, we introduce a facility location planning model for future US spaceports (SPFLP). Central considerations for the SPFLP include population density, space launch trajectories, and potential impacts to air traffic within the US National Airspace System (NAS). The SPFLP outputs a cost-optimal set of candidate locations for future spaceports while satisfying a range of operational constraints. By conducting sensitivity analyses on the SPFLP, we are able to examine differences in flight rerouting costs and optimal launch mission allocations. Our model and numerical experiments offer valuable insights for future spaceport site selection, contributing to the strategic development of commercial space transportation while keeping in mind the need to integrate these operations within the NAS.

math.OC↗

arXiv2023

SafeAR: Safe Algorithmic Recourse by Risk-Aware Policies

With the growing use of machine learning (ML) models in critical domains such as finance and healthcare, the need to offer recourse for those adversely affected by the decisions of ML models has become more important; individuals ought to be provided with recommendations on actions to take for improving their situation and thus receiving a favorable decision. Prior work on sequential algorithmic recourse -- which recommends a series of changes -- focuses on action feasibility and uses the proximity of feature changes to determine action costs. However, the uncertainties of feature changes and the risk of higher than average costs in recourse have not been considered. It is undesirable if a recourse could (with some probability) result in a worse situation from which recovery requires an extremely high cost. It is essential to incorporate risks when computing and evaluating recourse. We call the recourse computed with such risk considerations as Safe Algorithmic Recourse (SafeAR). The objective is to empower people to choose a recourse based on their risk tolerance. In this work, we discuss and show how existing recourse desiderata can fail to capture the risk of higher costs. We present a method to compute recourse policies that consider variability in cost and connect algorithmic recourse literature with risk-sensitive reinforcement learning. We also adopt measures "Value at Risk" and "Conditional Value at Risk" from the financial literature to summarize risk concisely. We apply our method to two real-world datasets and compare policies with different risk-aversion levels using risk measures and recourse desiderata (sparsity and proximity).

cs.LG↗

arXiv2023

Distributionally Robust Airport Ground Holding Problem under Wasserstein Ambiguity Sets

Ground Delay Programs (GDPs) mitigate demand-capacity imbalances by holding flights on the ground when an airport's arrival capacity is reduced, thereby reducing costly airborne holding. A central challenge is that day-to-day demand-capacity balancing relies on accurate predictions of airport capacities. However, these predictions are deeply uncertain: forecast errors, operational disruptions, and climate change-driven shifts in weather severity can induce distribution shifts in capacity outcomes. Thus, policies optimized for a single predicted distribution may be brittle out of sample. We address this challenge by developing a \emph{distributionally robust} framework for the single airport ground holding problem (dr-SAGHP). We also propose a method integrates Kelly's cutting plane method with the integer L-shaped method, and that is applicable more broadly to two-stage distributionally robust integer programs with relatively complete recourse and continuous second-stage decision variables. Our method includes a novel dual bisection and primal recovery algorithm that makes use of the structure of the distributionally robust integer program in order to quickly generate subgradients required by Kelly's cutting plane method. In computational experiments, our proposed algorithm delivers up to two orders-of-magnitude speedups compared to solving the convex reformulation directly, while maintaining negligible optimality gaps. We generate capacity scenarios via Gaussian process regression and evaluate out-of-sample performance by perturbing the posterior mean and variance. The numerical experiment results show that dr-SAGHP delivers significant out-of-sample gains under moderate-to-severe shifts, improving the resilience and effectiveness of GDP decision-making under capacity uncertainty.

math.OC↗

arXiv2023

Multiagent Inverse Reinforcement Learning via Theory of Mind Reasoning

We approach the problem of understanding how people interact with each other in collaborative settings, especially when individuals know little about their teammates, via Multiagent Inverse Reinforcement Learning (MIRL), where the goal is to infer the reward functions guiding the behavior of each individual given trajectories of a team's behavior during some task. Unlike current MIRL approaches, we do not assume that team members know each other's goals a priori; rather, that they collaborate by adapting to the goals of others perceived by observing their behavior, all while jointly performing a task. To address this problem, we propose a novel approach to MIRL via Theory of Mind (MIRL-ToM). For each agent, we first use ToM reasoning to estimate a posterior distribution over baseline reward profiles given their demonstrated behavior. We then perform MIRL via decentralized equilibrium by employing single-agent Maximum Entropy IRL to infer a reward function for each agent, where we simulate the behavior of other teammates according to the time-varying distribution over profiles. We evaluate our approach in a simulated 2-player search-and-rescue operation where the goal of the agents, playing different roles, is to search for and evacuate victims in the environment. Our results show that the choice of baseline profiles is paramount to the recovery of the ground-truth rewards, and that MIRL-ToM is able to recover the rewards used by agents interacting both with known and unknown teammates.

cs.AI↗

arXiv2022

Accelerated RRT* By Local Directional Visibility

RRT* is an efficient sampling-based motion planning algorithm. However, without taking advantages of accessible environment information, sampling-based algorithms usually result in sampling failures, generate useless nodes, and/or fail in exploring narrow passages. For this paper, in order to better utilize environment information and further improve searching efficiency, we proposed a novel approach to improve RRT* by 1) quantifying local knowledge of the obstacle configurations during neighbour rewiring in terms of directional visibility, 2) collecting environment information during searching, and 3) changing the sampling strategy biasing toward near-obstacle nodes after the first solution found. The proposed algorithm RRT* by Local Directional Visibility (RRT*-LDV) better utilizes local known information and innovates a weighted sampling strategy. The accelerated RRT*-LDV outperforms RRT* in convergence rate and success rate of finding narrow passages. A high Degree-Of-Freedom scenario is also experimented.

cs.RO↗

arXiv2022

Task Allocation with Load Management in Multi-Agent Teams

In operations of multi-agent teams ranging from homogeneous robot swarms to heterogeneous human-autonomy teams, unexpected events might occur. While efficiency of operation for multi-agent task allocation problems is the primary objective, it is essential that the decision-making framework is intelligent enough to manage unexpected task load with limited resources. Otherwise, operation effectiveness would drastically plummet with overloaded agents facing unforeseen risks. In this work, we present a decision-making framework for multi-agent teams to learn task allocation with the consideration of load management through decentralized reinforcement learning, where idling is encouraged and unnecessary resource usage is avoided. We illustrate the effect of load management on team performance and explore agent behaviors in example scenarios. Furthermore, a measure of agent importance in collaboration is developed to infer team resilience when facing handling potential overload situations.

cs.MA↗