Selected work

Representative Papers

TerraMind: Large-Scale Generative Multimodality for Earth Observation

Apr 15, 2025

To address the challenges of modeling Earth observation (EO) multimodal data—particularly the difficulty in jointly capturing fine-grained spatial details and high-level semantics—this paper introduces the first generative multimodal foundation model for EO supporting arbitrary modality-to-arbitrary modality translation. Methodologically, we propose a novel dual-scale (token-level + pixel-level) early-fusion pretraining paradigm, jointly trained on nine global geospatial modalities; we further introduce “Thinking-in-Modality” (TiM), a mechanism enabling dynamic in-modal sample augmentation during inference and fine-tuning. Our contributions include: (1) open-sourcing both the model weights and a high-quality, large-scale EO multimodal dataset; and (2) achieving state-of-the-art performance across standard benchmarks (e.g., PANGAEA), unifying cross-modal generation, semantic understanding, and spatial reasoning within a single framework, while significantly improving zero-shot and few-shot generalization capabilities.

1 citationsRead paper

Evaluating QAOA expectation values can be as hard as counting optimal solutions

Aug 11, 2026

This work investigates the computational complexity of evaluating expectation values in the Quantum Approximate Optimization Algorithm (QAOA) for the MaxCut problem. By employing deterministic polynomial-time Turing reductions, Laurent polynomial analysis, and #P-hardness proof techniques, it establishes that for circuit depth \( p \geq 2 \), computing exact or exponentially precise expectation values—and their gradients and Hessians—is #P-hard, even when restricted to a single two-body correlation term or a constrained parameter set. This result demonstrates that the complexity transition from \( p = 1 \) to \( p \geq 2 \) in QAOA transcends mere NP-hardness, ascending to the #P level associated with counting optimal solutions. Moreover, the reduction simultaneously recovers both the maximum cut value and the number of optimal solutions, thereby revealing a fundamental computational barrier inherent to this task.

0 citationsRead paper

SeFaR: Semantic Feature-aware Robustness Testing of Deep Neural Networks

Aug 10, 2026

This work addresses the insufficient robustness of deep neural networks in safety-critical scenarios caused by rare semantic shifts and the difficulty of verifying high-level semantic requirements. To tackle these challenges, the authors propose SeFaR, a framework that integrates diffusion models with vision-language models to generate realistic, semantically consistent, and diverse perturbations guided by natural language specifications and valid inputs. SeFaR employs a hierarchical concept model to systematically explore the feature space, incorporates domain knowledge through user-defined concepts, and leverages a feedback mechanism to identify requirement-irrelevant features that influence model decisions. This enables the generation of interpretable failure-inducing semantic concepts along with corresponding test samples. Experimental results demonstrate that SeFaR effectively uncovers model failures and accurately attributes them to specific semantic factors.

0 citationsRead paper

SPECTRA: Band-Routed Embedding and Stage-Wise LoRA for Cross-Sensor Fine-Tuning of Geospatial Foundation Models

Aug 03, 2026

This work addresses the challenges of spectral mismatch and high fine-tuning costs when adapting pretrained geospatial foundation models (GeoFMs) for Earth observation downstream tasks. To this end, we propose the SPECTRA framework, which introduces a Band-Routing Embedding (BRE) mechanism to effectively integrate all downstream spectral bands into the pretrained model’s input space. Additionally, we design a Stage-wise Transferability-aware Low-Rank Adaptation (ST-LoRA) strategy that dynamically allocates adaptation rank across network stages based on their transferability. Experiments across three GeoFM architectures and four segmentation datasets demonstrate that BRE substantially improves performance, while ST-LoRA outperforms both full fine-tuning and standard LoRA with significantly fewer trainable parameters. To our knowledge, this is the first approach to jointly optimize spectral alignment and parameter-efficient fine-tuning in geospatial foundation models.

0 citationsRead paper

A Taxonomy of Human-Robot Teamwork Requirements

Jul 29, 2026

This study addresses the lack of a unified requirements framework in current human-AI collaboration tasks, which hinders the design and evaluation of complex cooperative systems. By systematically reviewing academic literature, industry standards, and regulatory guidelines, this work proposes the first comprehensive requirements taxonomy encompassing six high-level categories and twenty-one subcategories. The framework integrates fragmented knowledge and clarifies critical dimensions such as information provision, relationship control, and decision support. Through an iterative process of requirement extraction, classification, and expert validation, the authors derived 361 requirements from 14 sources to construct the taxonomy and further demonstrated its applicability on an independent corpus containing 448 requirements. The resulting framework received endorsement from five domain experts.

0 citationsRead paper

Recent publications

Latest Papers

Evaluating QAOA expectation values can be as hard as counting optimal solutions

Aug 11, 2026

This work investigates the computational complexity of evaluating expectation values in the Quantum Approximate Optimization Algorithm (QAOA) for the MaxCut problem. By employing deterministic polynomial-time Turing reductions, Laurent polynomial analysis, and #P-hardness proof techniques, it establishes that for circuit depth \( p \geq 2 \), computing exact or exponentially precise expectation values—and their gradients and Hessians—is #P-hard, even when restricted to a single two-body correlation term or a constrained parameter set. This result demonstrates that the complexity transition from \( p = 1 \) to \( p \geq 2 \) in QAOA transcends mere NP-hardness, ascending to the #P level associated with counting optimal solutions. Moreover, the reduction simultaneously recovers both the maximum cut value and the number of optimal solutions, thereby revealing a fundamental computational barrier inherent to this task.

0 citationsRead paper

SeFaR: Semantic Feature-aware Robustness Testing of Deep Neural Networks

Aug 10, 2026

This work addresses the insufficient robustness of deep neural networks in safety-critical scenarios caused by rare semantic shifts and the difficulty of verifying high-level semantic requirements. To tackle these challenges, the authors propose SeFaR, a framework that integrates diffusion models with vision-language models to generate realistic, semantically consistent, and diverse perturbations guided by natural language specifications and valid inputs. SeFaR employs a hierarchical concept model to systematically explore the feature space, incorporates domain knowledge through user-defined concepts, and leverages a feedback mechanism to identify requirement-irrelevant features that influence model decisions. This enables the generation of interpretable failure-inducing semantic concepts along with corresponding test samples. Experimental results demonstrate that SeFaR effectively uncovers model failures and accurately attributes them to specific semantic factors.

0 citationsRead paper

SPECTRA: Band-Routed Embedding and Stage-Wise LoRA for Cross-Sensor Fine-Tuning of Geospatial Foundation Models

Aug 03, 2026

This work addresses the challenges of spectral mismatch and high fine-tuning costs when adapting pretrained geospatial foundation models (GeoFMs) for Earth observation downstream tasks. To this end, we propose the SPECTRA framework, which introduces a Band-Routing Embedding (BRE) mechanism to effectively integrate all downstream spectral bands into the pretrained model’s input space. Additionally, we design a Stage-wise Transferability-aware Low-Rank Adaptation (ST-LoRA) strategy that dynamically allocates adaptation rank across network stages based on their transferability. Experiments across three GeoFM architectures and four segmentation datasets demonstrate that BRE substantially improves performance, while ST-LoRA outperforms both full fine-tuning and standard LoRA with significantly fewer trainable parameters. To our knowledge, this is the first approach to jointly optimize spectral alignment and parameter-efficient fine-tuning in geospatial foundation models.

0 citationsRead paper

A Taxonomy of Human-Robot Teamwork Requirements

Jul 29, 2026

This study addresses the lack of a unified requirements framework in current human-AI collaboration tasks, which hinders the design and evaluation of complex cooperative systems. By systematically reviewing academic literature, industry standards, and regulatory guidelines, this work proposes the first comprehensive requirements taxonomy encompassing six high-level categories and twenty-one subcategories. The framework integrates fragmented knowledge and clarifies critical dimensions such as information provision, relationship control, and decision support. Through an iterative process of requirement extraction, classification, and expert validation, the authors derived 361 requirements from 14 sources to construct the taxonomy and further demonstrated its applicability on an independent corpus containing 448 requirements. The resulting framework received endorsement from five domain experts.

0 citationsRead paper

HDRL Staff Strategy Meeting Report

Jul 07, 2026

This study addresses the strategic challenges confronting HDRL across key domains—including metadata management, user experience, infrastructure, impact assessment, and community sustainability—by designing and implementing a large-scale participatory strategic planning process. Through an organization-wide convening, the initiative integrated over 30 lightning talks and more than 50 discussion topics contributed by 40+ members, leveraging collaborative deliberation, collective voting (324 votes in total), and thematic clustering analysis to systematically distill seven core strategic themes. The approach innovatively combines multidimensional stakeholder engagement with data-driven decision-making, culminating in an actionable strategic roadmap that provides HDRL with a clear, consensus-based vision for its modernization and long-term development.

0 citationsRead paper