Mark Crowley

recent highlighted publications

1. A novel soil moisture retrieval method via combining radiative transfer model and machine learning

   Yurun Chen, Cheng Tong, Josh Qixuan Sun, Yulin Shangguan, Xiaodong Deng, Mark Crowley, Hongquan Wang, Yang Ye, Haijun Bao, and Ruqi Huang.

   Remote Sensing of Environment. 338, 2026.

   Abs URL

   Soil moisture (SM) is a key variable in the global water cycle, and passive microwave remote sensing is widely used for large-scale SM estimation. Although radiative transfer models (RTM) and machine learning (ML) approaches have advanced SM retrieval, interpretable methods that provide a direct, inspectable mapping from satellite observations to SM remain scarce. In this study, we present an initial attempt to develop such an interpretable retrieval framework by fusing the physical principles of the RTM with the mathematical formulation capability of the Kolmogorov–Arnold Network (KAN). The framework proceeds in two stages: first, a physically consistent training dataset is generated by the RTM that links SM with brightness temperature (TB), surface temperature (ST), and vegetation optical depth (VOD), thereby preserving key physical dependencies; second, KAN is trained on this dataset to extract explicit mathematical expressions that relate TB, ST and VOD to SM for different land-cover types. These derived expressions are applied to SMAP observations to produce global daily SM estimates for 2015–2023. Validation against in situ soil moisture measurements from global networks, including the International Soil Moisture Network (ISMN) and the Qinghai Lake Basin dense network (QLB-NET), shows that the KAN retrieval achieves an average correlation coefficient R = 0.64 and unbiased RMSE (ubRMSE) = 0.07 m3/m3, comparable to the SMAP Level-3 product (R = 0.65, ubRMSE =0.06 m3/m3) in capturing broad spatial patterns and seasonal dynamics. Beyond enabling accurate large-scale estimates, the explicit formula extracted by KAN clarifies how TB, ST, and VOD influence SM and provides a compact, direct retrieval expression that requires neither KAN retraining nor the iterative solving typical of traditional RTM approaches, thereby improving reproducibility and operational efficiency. This interpretable fusion of physical modeling and data-driven representation offers a novel pathway for advancing quantitative remote sensing retrievals and invites further refinement and operational evaluation.

2. Toward Virtuous Reinforcement Learning: A Critique and Roadmap

   Majid Ghasemi, and Mark Crowley

   Singapore. Jan, 2026.

   Abs arXiv PDF

   This paper critiques common patterns in machine ethics for Reinforcement Learning (RL) and argues for a virtue focused alternative. We highlight two recurring limitations in much of the current literature: (i) rule based (deontological) methods that encode duties as constraints or shields often struggle under ambiguity and nonstationarity and do not cultivate lasting habits, and (ii) many reward based approaches, especially single objective RL, implicitly compress diverse moral considerations into a single scalar signal, which can obscure trade offs and invite proxy gaming in practice. We instead treat ethics as policy level dispositions, that is, relatively stable habits that hold up when incentives, partners, or contexts change. This shifts evaluation beyond rule checks or scalar returns toward trait summaries, durability under interventions, and explicit reporting of moral trade offs. Our roadmap combines four components: (1) social learning in multi agent RL to acquire virtue like patterns from imperfect but normatively informed exemplars; (2) multi objective and constrained formulations that preserve value conflicts and incorporate risk aware criteria to guard against harm; (3) affinity based regularization toward updateable virtue priors that support trait like stability under distribution shift while allowing norms to evolve; and (4) operationalizing diverse ethical traditions as practical control signals, making explicit the value and cultural assumptions that shape ethical RL benchmarks.

3. Learning When to Observe: A Frugal Reinforcement Learning Framework for a High-Cost World

   Colin Bellinger, Isaac Tamblyn, and Mark Crowley

   SPRINGER INTERNATIONAL PUBLISHING AG, GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND. 2025.

   Abs

   Reinforcement learning (RL) has been shown to learn sophisticated control policies for complex tasks including games, robotics, heating and cooling systems and text generation. The action-perception cycle in RL, however, generally assumes that a measurement of the state of the environment is available at each time step without a cost. In applications such as materials design, deep-sea and planetary robot exploration and medicine, however, there can be a high cost associated with measuring, or even approximating, the state of the environment. In this paper, we survey the recently growing literature that adopts the perspective that an RL agent might not need, or even want, a costly measurement at each time step. Within this context, we propose the Deep Dynamic Multi-Step Observationless Agent (DMSOA), contrast it with the literature and empirically evaluate it on OpenAI gym and Atari Pong environments. Our results, show that DMSOA learns a better policy with fewer decision steps and measurements than the considered alternative from the literature. The corresponding code is available at: https://github.com/cbellinger27/Learning-when-to-observe-in-RL.

4. Dynamic programming with incomplete information to overcomenavigational uncertainty in POMDPs

   Chris Beeler, Xinkai Li, Colin Bellinger, Mark Crowley, Maia Fraser, and Isaac Tamblyn.

   In Proceedings of the Canadian Conference on Artificial Intelligence. Canadian Artificial Intelligence Association (CAIAC), Guelph, Ontario, Canada. May, 2024.

   Abs PDF URL

   Using a generalizable novel nautical navigation environment, we show how dynamic programming can be used when only incomplete information about a partially observed Markov decision process (POMDP) is known. By incorporating uncertainty into our model, we show that navigation policies can be constructed that maintain safety, outperforming the baseline performance of traditional dynamic programming for Markov decision processes (MDPs). Adding in controlled sensing methods, we show that these policies can also lower measurement costs at the same time.

5. Causal2

   [preprint] Implicit Causal Representation Learning via Switchable Mechanisms

   Shayan Shirahmad Gale Bagi, Zahra Gharaee, Oliver Schulte, and Mark Crowley

   In Arxiv Preprint. 2024.

   arXiv PDF