2025
11.
Oishee Bintey Hoque, Abhijin Adiga, Aniruddha Adiga, Siddharth Chaudhary, Madhav V. Marathe, S. S. Ravi, Kirti Rajagopalan, Amanda Wilson,; Samarth Swarup
IGraSS: Learning to Identify Infrastructure Networks from Satellite Imagery by Iterative Graph-constrained Semantic Segmentation Proceedings Article
In: 2025 IEEE International Geoscience and Remote Sensing Symposium 2025.
Abstract | Links | BibTeX | Tags: Infrastructure Networks, Satellite Imagery, Semantic Segmentation
@inproceedings{nokey,
title = {IGraSS: Learning to Identify Infrastructure Networks from Satellite Imagery by Iterative Graph-constrained Semantic Segmentation},
author = {Oishee Bintey Hoque, Abhijin Adiga, Aniruddha Adiga, Siddharth Chaudhary, Madhav V. Marathe, S. S. Ravi, Kirti Rajagopalan, Amanda Wilson, and Samarth Swarup},
url = {https://arxiv.org/abs/2506.08137},
doi = { https://doi.org/10.48550/arXiv.2506.08137},
year = {2025},
date = {2025-08-08},
urldate = {2025-08-08},
organization = {2025 IEEE International Geoscience and Remote Sensing Symposium},
abstract = {Accurate canal network mapping is essential for water management, including irrigation planning and infrastructure maintenance. State-of-the-art semantic segmentation models for infrastructure mapping, such as roads, rely on large, wellannotated remote sensing datasets. However, incomplete or inadequate ground truth can hinder these learning approaches. Many infrastructure networks have graph-level properties such as reachability to a source (like canals) or connectivity (roads) that can be leveraged to improve these existing ground truth. This paper develops a novel iterative framework IGraSS, combining a semantic segmentation module\textemdashincorporating RGB and additional modalities (NDWI, DEM)\textemdashwith a graph-based ground-truth refinement module. The segmentation module processes satellite imagery patches, while the refinement module operates on the entire data viewing the infrastructure network as a graph. Experiments show that IGraSS reduces unreachable canal segments from 18% to 3%, and training with refined ground truth significantly improves canal identification. IGraSS serves as a robust framework for both refining noisy ground truth and mapping canal networks from remote sensing imagery. We also demonstrate the effectiveness and generalizability of IGraSS using road networks as an example, applying a different graph-theoretic constraint to complete road networks.},
keywords = {Infrastructure Networks, Satellite Imagery, Semantic Segmentation},
pubstate = {published},
tppubtype = {inproceedings}
}
Accurate canal network mapping is essential for water management, including irrigation planning and infrastructure maintenance. State-of-the-art semantic segmentation models for infrastructure mapping, such as roads, rely on large, wellannotated remote sensing datasets. However, incomplete or inadequate ground truth can hinder these learning approaches. Many infrastructure networks have graph-level properties such as reachability to a source (like canals) or connectivity (roads) that can be leveraged to improve these existing ground truth. This paper develops a novel iterative framework IGraSS, combining a semantic segmentation module—incorporating RGB and additional modalities (NDWI, DEM)—with a graph-based ground-truth refinement module. The segmentation module processes satellite imagery patches, while the refinement module operates on the entire data viewing the infrastructure network as a graph. Experiments show that IGraSS reduces unreachable canal segments from 18% to 3%, and training with refined ground truth significantly improves canal identification. IGraSS serves as a robust framework for both refining noisy ground truth and mapping canal networks from remote sensing imagery. We also demonstrate the effectiveness and generalizability of IGraSS using road networks as an example, applying a different graph-theoretic constraint to complete road networks.
12.
Nibir Chandra Mandal, Oishee Bintey Hoque, Abhijin Adiga, Samarth Swarup, Mandy Wilson, Lu Feng, Yangfeng Ji, Miaomiao Zhang, Geoffrey Fox, Madhav Marathe
IrrMap: A Large-Scale Comprehensive Dataset for Irrigation Method Mapping Proceedings Article
In: 2025 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 2025.
Abstract | Links | BibTeX | Tags: AI, Irrigation, Mapping
@inproceedings{nokey,
title = {IrrMap: A Large-Scale Comprehensive Dataset for Irrigation Method Mapping},
author = {Nibir Chandra Mandal, Oishee Bintey Hoque, Abhijin Adiga, Samarth Swarup, Mandy Wilson, Lu Feng, Yangfeng Ji, Miaomiao Zhang, Geoffrey Fox, Madhav Marathe},
doi = { https://doi.org/10.48550/arXiv.2505.08273},
year = {2025},
date = {2025-08-07},
urldate = {2025-08-07},
organization = {2025 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
abstract = {We introduce IrrMap, the first large-scale dataset (1.1 million patches) for irrigation method mapping across regions. IrrMap consists of multi-resolution satellite imagery from LandSat and Sentinel, along with key auxiliary data such as crop type, land use, and vegetation indices. The dataset spans 1,687,899 farms and 14,117,330 acres across multiple western U.S. states from 2013 to 2023, providing a rich and diverse foundation for irrigation analysis and ensuring geospatial alignment and quality control. The dataset is ML-ready, with standardized 224×224 GeoTIFF patches, the multiple input data layers, carefully chosen train-test-split data, and accompanying dataloaders for seamless deep learning model training and benchmarking in irrigation mapping. The dataset is also accompanied by a complete pipeline for dataset generation, enabling researchers to extend IrrMap to new regions for irrigation data collection or adapt it with minimal effort for other similar applications in agricultural and geospatial analysis. We also analyze the irrigation method distribution across crop groups, spatial irrigation patterns (using Shannon diversity indices), and irrigated area variations for both LandSat and Sentinel, providing insights into regional and resolution-based differences. To promote further exploration, we openly release IrrMap, along with the derived datasets, benchmark models, and pipeline code, through a GitHub repository: https:// github.com/ Nibir088/IrrMap and Data repository: https:// huggingface.co/ Nibir/IrrMap, providing comprehensive documentation and implementation details.},
keywords = {AI, Irrigation, Mapping},
pubstate = {published},
tppubtype = {inproceedings}
}
We introduce IrrMap, the first large-scale dataset (1.1 million patches) for irrigation method mapping across regions. IrrMap consists of multi-resolution satellite imagery from LandSat and Sentinel, along with key auxiliary data such as crop type, land use, and vegetation indices. The dataset spans 1,687,899 farms and 14,117,330 acres across multiple western U.S. states from 2013 to 2023, providing a rich and diverse foundation for irrigation analysis and ensuring geospatial alignment and quality control. The dataset is ML-ready, with standardized 224×224 GeoTIFF patches, the multiple input data layers, carefully chosen train-test-split data, and accompanying dataloaders for seamless deep learning model training and benchmarking in irrigation mapping. The dataset is also accompanied by a complete pipeline for dataset generation, enabling researchers to extend IrrMap to new regions for irrigation data collection or adapt it with minimal effort for other similar applications in agricultural and geospatial analysis. We also analyze the irrigation method distribution across crop groups, spatial irrigation patterns (using Shannon diversity indices), and irrigated area variations for both LandSat and Sentinel, providing insights into regional and resolution-based differences. To promote further exploration, we openly release IrrMap, along with the derived datasets, benchmark models, and pipeline code, through a GitHub repository: https:// github.com/ Nibir088/IrrMap and Data repository: https:// huggingface.co/ Nibir/IrrMap, providing comprehensive documentation and implementation details.
13.
Hooman Shahrokhi; Devjeet Raj Roy; Yan Yan; Venera Arnaoudova; Jana Doppa
Conformal Prediction Sets for Deep Generative Models via Reduction to Conformal Regression Proceedings Article
In: Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence, pp. 3718–3748, PMLR, 2025, (ISSN: 2640-3498).
Abstract | Links | BibTeX | Tags: AI
@inproceedings{shahrokhi_conformal_2025,
title = {Conformal Prediction Sets for Deep Generative Models via Reduction to Conformal Regression},
author = {Hooman Shahrokhi and Devjeet Raj Roy and Yan Yan and Venera Arnaoudova and Jana Doppa},
url = {https://proceedings.mlr.press/v286/shahrokhi25a.html},
year = {2025},
date = {2025-07-01},
urldate = {2025-07-01},
booktitle = {Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence},
pages = {3718\textendash3748},
publisher = {PMLR},
abstract = {We consider the problem of generating valid and small prediction sets by sampling outputs (e.g., software code and natural language text) from a black-box deep generative model for a given input (e.g., textual prompt). The validity of a prediction set is determined by a user-defined binary admissibility function depending on the target application. For example, requiring at least one program in the set to pass all test cases in code generation application. To address this problem, we develop a simple and effective conformal inference algorithm referred to as textbackslashem Generative Prediction Sets (GPS). Given a set of calibration examples and black-box access to a deep generative model, GPS can generate prediction sets with provable guarantees. The key insight behind GPS is to exploit the inherent structure within the distribution over the minimum number of samples needed to obtain an admissible output to develop a simple conformal regression approach over the minimum number of samples. Unlike prior work , the sets generated by GPS do not require iterative sampling at test time, while maintaining strict marginal coverage guarantees. Experiments on multiple datasets for code and math word problems using different large language models demonstrate the efficacy of GPS over state-of-the-art methods.},
note = {ISSN: 2640-3498},
keywords = {AI},
pubstate = {published},
tppubtype = {inproceedings}
}
We consider the problem of generating valid and small prediction sets by sampling outputs (e.g., software code and natural language text) from a black-box deep generative model for a given input (e.g., textual prompt). The validity of a prediction set is determined by a user-defined binary admissibility function depending on the target application. For example, requiring at least one program in the set to pass all test cases in code generation application. To address this problem, we develop a simple and effective conformal inference algorithm referred to as textbackslashem Generative Prediction Sets (GPS). Given a set of calibration examples and black-box access to a deep generative model, GPS can generate prediction sets with provable guarantees. The key insight behind GPS is to exploit the inherent structure within the distribution over the minimum number of samples needed to obtain an admissible output to develop a simple conformal regression approach over the minimum number of samples. Unlike prior work , the sets generated by GPS do not require iterative sampling at test time, while maintaining strict marginal coverage guarantees. Experiments on multiple datasets for code and math word problems using different large language models demonstrate the efficacy of GPS over state-of-the-art methods.
14.
John T. Abatzoglou; Charles A. Young; Vishal K. Mehta; Joshua H. Viers; Katherine C. Hegewisch
Predicting California Water-Year Types Using Seasonal Climate Forecasts Journal Article
In: 2025, (Section: Journal of Applied Meteorology and Climatology).
Abstract | Links | BibTeX | Tags: AI, Water
@article{abatzoglou_predicting_2025,
title = {Predicting California Water-Year Types Using Seasonal Climate Forecasts},
author = {John T. Abatzoglou and Charles A. Young and Vishal K. Mehta and Joshua H. Viers and Katherine C. Hegewisch},
url = {https://journals.ametsoc.org/view/journals/apme/64/8/JAMC-D-24-0244.1.xml},
doi = {10.1175/JAMC-D-24-0244.1},
year = {2025},
date = {2025-07-01},
urldate = {2025-07-01},
abstract = {California’s water management is confounded by large interannual variability in water availability inherent in Mediterranean climates and competing water demands in the state. Hydrologic outlooks during late winter and spring are an important consideration that water managers use in water allocation decisions. While hydrologic outlooks are informed by initial conditions, particularly in snowmelt-dominated systems such as California’s Sierra Nevada, there is potential for seasonal climate forecasts to improve such outlooks. California uses a standard water-year typology based on indices of runoff for the Sacramento and San Joaquin Rivers which form the basis of state surface water allocation decisions. We present a simple intuitive model that uses November\textendashMarch watershed average precipitation and temperature in each river basin that explains approximately 90% of the variance in water-year type indices. We then evaluate the utility of these models to forecast water-year types by appending observational temperature and precipitation data with downscaled seasonal climate forecasts from the North American Multimodel Ensemble (NMME). We demonstrate that seasonal climate forecasts augment forecasts based exclusively on initial conditions (e.g., Sierra snowpack). For example, NMME-informed forecasts in January for critical or dry water-year types had an accuracy of 50%\textemdashcomparable to the skill of forecasts based solely on initial conditions 1\textendash2 months later. While California’s hydroclimate is renowned for its unpredictability, results suggest that NMME-informed forecasts may provide additional lead time for water managers to proactively implement water management strategies in anticipation of dry conditions. Significance Statement Water resources in California are wildly variable from year to year resulting in challenges for water managers and users. Improved hydroclimate forecasts can enable proactive measures to secure water for multiple uses. Seasonal forecasts of winter precipitation are notoriously poor in California’s Sierra Nevada, which provides much of the state’s water resources. However, here we show that incorporating seasonal temperature and precipitation forecasts alongside observations adds skill to water-year type outlooks in the Sacramento and San Joaquin River basins. Such forecasts can better inform surface water allocation decisions to minimize the economic and ecological impacts of the state’s variable hydroclimate.},
note = {Section: Journal of Applied Meteorology and Climatology},
keywords = {AI, Water},
pubstate = {published},
tppubtype = {article}
}
California’s water management is confounded by large interannual variability in water availability inherent in Mediterranean climates and competing water demands in the state. Hydrologic outlooks during late winter and spring are an important consideration that water managers use in water allocation decisions. While hydrologic outlooks are informed by initial conditions, particularly in snowmelt-dominated systems such as California’s Sierra Nevada, there is potential for seasonal climate forecasts to improve such outlooks. California uses a standard water-year typology based on indices of runoff for the Sacramento and San Joaquin Rivers which form the basis of state surface water allocation decisions. We present a simple intuitive model that uses November–March watershed average precipitation and temperature in each river basin that explains approximately 90% of the variance in water-year type indices. We then evaluate the utility of these models to forecast water-year types by appending observational temperature and precipitation data with downscaled seasonal climate forecasts from the North American Multimodel Ensemble (NMME). We demonstrate that seasonal climate forecasts augment forecasts based exclusively on initial conditions (e.g., Sierra snowpack). For example, NMME-informed forecasts in January for critical or dry water-year types had an accuracy of 50%—comparable to the skill of forecasts based solely on initial conditions 1–2 months later. While California’s hydroclimate is renowned for its unpredictability, results suggest that NMME-informed forecasts may provide additional lead time for water managers to proactively implement water management strategies in anticipation of dry conditions. Significance Statement Water resources in California are wildly variable from year to year resulting in challenges for water managers and users. Improved hydroclimate forecasts can enable proactive measures to secure water for multiple uses. Seasonal forecasts of winter precipitation are notoriously poor in California’s Sierra Nevada, which provides much of the state’s water resources. However, here we show that incorporating seasonal temperature and precipitation forecasts alongside observations adds skill to water-year type outlooks in the Sacramento and San Joaquin River basins. Such forecasts can better inform surface water allocation decisions to minimize the economic and ecological impacts of the state’s variable hydroclimate.
15.
Soumiki Chattopadhyay
Systematizing inclusive design in MOSIP : an experience report Technical Report
2025, (Publisher: Oregon State University).
Links | BibTeX | Tags: AI, Humans
@techreport{chattopadhyay_systematizing_nodateb,
title = {Systematizing inclusive design in MOSIP : an experience report},
author = {Soumiki Chattopadhyay},
url = {https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/0g354q07x},
year = {2025},
date = {2025-06-10},
note = {Publisher: Oregon State University},
keywords = {AI, Humans},
pubstate = {published},
tppubtype = {techreport}
}
16.
Ranjan Sapkota; Marco Flores-Calero; Rizwan Qureshi; Chetan Badgujar; Upesh Nepal; Alwin Poulose; Peter Zeno; Uday Bhanu Prakash Vaddevolu; Sheheryar Khan; Maged Shoman; Hong Yan; Manoj Karkee
YOLO advances to its genesis: a decadal and comprehensive review of the You Only Look Once (YOLO) series Journal Article
In: Artificial Intelligence Review, vol. 58, no. 9, pp. 274, 2025, ISSN: 1573-7462.
Abstract | Links | BibTeX | Tags: Agriculture, Artificial intelligence, Autonomous vehicles, CNN, Computer vision, Deep learning, Healthcare and medical imaging, Industrial manufacturing, Real-time object detection, Surveillance, Traffic safety, YOLO, YOLO configurations, YOLOv1 to YOLOv12, You Only Look Once
@article{sapkota_yolo_2025,
title = {YOLO advances to its genesis: a decadal and comprehensive review of the You Only Look Once (YOLO) series},
author = {Ranjan Sapkota and Marco Flores-Calero and Rizwan Qureshi and Chetan Badgujar and Upesh Nepal and Alwin Poulose and Peter Zeno and Uday Bhanu Prakash Vaddevolu and Sheheryar Khan and Maged Shoman and Hong Yan and Manoj Karkee},
url = {https://doi.org/10.1007/s10462-025-11253-3},
doi = {10.1007/s10462-025-11253-3},
issn = {1573-7462},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-01},
journal = {Artificial Intelligence Review},
volume = {58},
number = {9},
pages = {274},
abstract = {This review systematically examines the progression of the You Only Look Once (YOLO) object detection algorithms from YOLOv1 to the recently unveiled YOLOv12. Employing a reverse chronological analysis, this study examines the advancements introduced by YOLO algorithms, beginning with YOLOv12 and progressing through YOLO11 (or YOLOv11), YOLOv10, YOLOv9, YOLOv8, and subsequent versions to explore each version’s contributions to enhancing speed, detection accuracy, and computational efficiency in real-time object detection. Additionally, this study reviews the alternative versions derived from YOLO architectural advancements of YOLO-NAS, YOLO-X, YOLO-R, DAMO-YOLO, and Gold-YOLO. Moreover, the study highlights the transformative impact of YOLO models across five critical application areas: autonomous vehicles and traffic safety, healthcare and medical imaging, industrial manufacturing, surveillance and security, and agriculture. By detailing the incremental technological advancements in subsequent YOLO versions, this review chronicles the evolution of YOLO, and discusses the challenges and limitations in each of the earlier versions. The evolution signifies a path towards integrating YOLO with multimodal, context-aware, and Artificial General Intelligence (AGI) systems for the next YOLO decade, promising significant implications for future developments in AI-driven applications.},
keywords = {Agriculture, Artificial intelligence, Autonomous vehicles, CNN, Computer vision, Deep learning, Healthcare and medical imaging, Industrial manufacturing, Real-time object detection, Surveillance, Traffic safety, YOLO, YOLO configurations, YOLOv1 to YOLOv12, You Only Look Once},
pubstate = {published},
tppubtype = {article}
}
This review systematically examines the progression of the You Only Look Once (YOLO) object detection algorithms from YOLOv1 to the recently unveiled YOLOv12. Employing a reverse chronological analysis, this study examines the advancements introduced by YOLO algorithms, beginning with YOLOv12 and progressing through YOLO11 (or YOLOv11), YOLOv10, YOLOv9, YOLOv8, and subsequent versions to explore each version’s contributions to enhancing speed, detection accuracy, and computational efficiency in real-time object detection. Additionally, this study reviews the alternative versions derived from YOLO architectural advancements of YOLO-NAS, YOLO-X, YOLO-R, DAMO-YOLO, and Gold-YOLO. Moreover, the study highlights the transformative impact of YOLO models across five critical application areas: autonomous vehicles and traffic safety, healthcare and medical imaging, industrial manufacturing, surveillance and security, and agriculture. By detailing the incremental technological advancements in subsequent YOLO versions, this review chronicles the evolution of YOLO, and discusses the challenges and limitations in each of the earlier versions. The evolution signifies a path towards integrating YOLO with multimodal, context-aware, and Artificial General Intelligence (AGI) systems for the next YOLO decade, promising significant implications for future developments in AI-driven applications.
17.
Yassine Chemingui; Aryan Deshwal; Honghao Wei; Alan Fern; Janardhan Rao Doppa
Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning Proceedings Article
In: pp. 15722-15730, Proceedings of the AAAI Conference on Artificial Intelligence, 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{chemingui_constraint-adaptive_2024,
title = {Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning},
author = {Yassine Chemingui and Aryan Deshwal and Honghao Wei and Alan Fern and Janardhan Rao Doppa},
url = {https://ojs.aaai.org/index.php/AAAI/article/view/33726},
doi = {10.1609/aaai.v39i15.33726},
year = {2025},
date = {2025-04-11},
urldate = {2025-04-11},
volume = {39},
number = {15},
pages = {15722-15730},
publisher = {Proceedings of the AAAI Conference on Artificial Intelligence},
abstract = {Offline safe reinforcement learning (OSRL) involves learning a decision-making policy to maximize rewards from a fixed batch of training data to satisfy pre-defined safety constraints. However, adapting to varying safety constraints during deployment without retraining remains an under-explored challenge. To address this challenge, we introduce constraint-adaptive policy switching (CAPS), a wrapper framework around existing offline RL algorithms. During training, CAPS uses offline data to learn multiple policies with a shared representation that optimize different reward and cost trade-offs. During testing, CAPS switches between those policies by selecting at each state the policy that maximizes future rewards among those that satisfy the current cost constraint. Our experiments on 38 tasks from the DSRL benchmark demonstrate that CAPS consistently outperforms existing methods, establishing a strong wrapper-based baseline for OSRL. The code is publicly available at https://github.com/yassineCh/CAPS.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Offline safe reinforcement learning (OSRL) involves learning a decision-making policy to maximize rewards from a fixed batch of training data to satisfy pre-defined safety constraints. However, adapting to varying safety constraints during deployment without retraining remains an under-explored challenge. To address this challenge, we introduce constraint-adaptive policy switching (CAPS), a wrapper framework around existing offline RL algorithms. During training, CAPS uses offline data to learn multiple policies with a shared representation that optimize different reward and cost trade-offs. During testing, CAPS switches between those policies by selecting at each state the policy that maximizes future rewards among those that satisfy the current cost constraint. Our experiments on 38 tasks from the DSRL benchmark demonstrate that CAPS consistently outperforms existing methods, establishing a strong wrapper-based baseline for OSRL. The code is publicly available at https://github.com/yassineCh/CAPS.
18.
Alexis Fox; Samarth Swarup; Abhijin Adiga
A Unifying Information-theoretic Perspective on Evaluating Generative Models Proceedings Article
In: pp. 16630–16638, 2025, ISSN: 2374-3468.
Abstract | Links | BibTeX | Tags: AI, Water
@inproceedings{fox_unifying_2025,
title = {A Unifying Information-theoretic Perspective on Evaluating Generative Models},
author = {Alexis Fox and Samarth Swarup and Abhijin Adiga},
url = {https://ojs.aaai.org/index.php/AAAI/article/view/33827},
doi = {10.1609/aaai.v39i16.33827},
issn = {2374-3468},
year = {2025},
date = {2025-04-01},
urldate = {2025-04-01},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {39},
number = {16},
pages = {16630\textendash16638},
abstract = {Considering the difficulty of interpreting generative model output, there is significant current research focused on determining meaningful evaluation metrics. Several recent approaches utilize "precision" and "recall," borrowed from the classification domain, to individually quantify the output fidelity (realism) and output diversity (representation of the real data variation), respectively. With the increase in metric proposals, there is a need for a unifying perspective, allowing for easier comparison and clearer explanation of their benefits and drawbacks. To this end, we unify a class of kth-nearest neighbors (kNN)-based metrics under an information-theoretic lens using approaches from kNN density estimation. Additionally, we propose a tri-dimensional metric composed of Precision Cross-Entropy (PCE), Recall Cross-Entropy (RCE), and Recall Entropy (RE), which separately measure fidelity and two distinct aspects of diversity, inter- and intra-class. Our domain-agnostic metric, derived from the information-theoretic concepts of entropy and cross-entropy, can be dissected for both sample- and mode-level analysis. Our detailed experimental results demonstrate the sensitivity of our metric components to their respective qualities and reveal undesirable behaviors of other metrics.},
keywords = {AI, Water},
pubstate = {published},
tppubtype = {inproceedings}
}
Considering the difficulty of interpreting generative model output, there is significant current research focused on determining meaningful evaluation metrics. Several recent approaches utilize "precision" and "recall," borrowed from the classification domain, to individually quantify the output fidelity (realism) and output diversity (representation of the real data variation), respectively. With the increase in metric proposals, there is a need for a unifying perspective, allowing for easier comparison and clearer explanation of their benefits and drawbacks. To this end, we unify a class of kth-nearest neighbors (kNN)-based metrics under an information-theoretic lens using approaches from kNN density estimation. Additionally, we propose a tri-dimensional metric composed of Precision Cross-Entropy (PCE), Recall Cross-Entropy (RCE), and Recall Entropy (RE), which separately measure fidelity and two distinct aspects of diversity, inter- and intra-class. Our domain-agnostic metric, derived from the information-theoretic concepts of entropy and cross-entropy, can be dissected for both sample- and mode-level analysis. Our detailed experimental results demonstrate the sensitivity of our metric components to their respective qualities and reveal undesirable behaviors of other metrics.
19.
Kristen Goebel; Paola Pesantez-Cabrera; Markus Keller; Alan Fern
Transfer Learning via Auxiliary Labels with Application to Cold-Hardiness Prediction Working paper
2025, (arXiv:2504.13142 [cs]).
Abstract | Links | BibTeX | Tags:
@workingpaper{goebel_transfer_2025,
title = {Transfer Learning via Auxiliary Labels with Application to Cold-Hardiness Prediction},
author = {Kristen Goebel and Paola Pesantez-Cabrera and Markus Keller and Alan Fern},
url = {http://arxiv.org/abs/2504.13142},
doi = {10.48550/arXiv.2504.13142},
year = {2025},
date = {2025-04-01},
urldate = {2025-04-01},
publisher = {arXiv},
abstract = {Cold temperatures can cause significant frost damage to fruit crops depending on their resilience, or cold hardiness, which changes throughout the dormancy season. This has led to the development of predictive cold-hardiness models, which help farmers decide when to deploy expensive frost-mitigation measures. Unfortunately, cold-hardiness data for model training is only available for some fruit cultivars due to the need for specialized equipment and expertise. Rather, farmers often do have years of phenological data (e.g. date of budbreak) that they regularly collect for their crops. In this work, we introduce a new transfer-learning framework, Transfer via Auxiliary Labels (TAL), that allows farmers to leverage the phenological data to produce more accurate cold-hardiness predictions, even when no cold-hardiness data is available for their specific crop. The framework assumes a set of source tasks (cultivars) where each has associated primary labels (cold hardiness) and auxiliary labels (phenology). However, the target task (new cultivar) is assumed to only have the auxiliary labels. The goal of TAL is to predict primary labels for the target task via transfer from the source tasks. Surprisingly, despite the vast literature on transfer learning, to our knowledge, the TAL formulation has not been previously addressed. Thus, we propose several new TAL approaches based on model selection and averaging that can leverage recent deep multi-task models for cold-hardiness prediction. Our results on real-world cold-hardiness and phenological data for multiple grape cultivars demonstrate that TAL can leverage the phenological data to improve cold-hardiness predictions in the absence of cold-hardiness data.},
note = {arXiv:2504.13142 [cs]},
keywords = {},
pubstate = {published},
tppubtype = {workingpaper}
}
Cold temperatures can cause significant frost damage to fruit crops depending on their resilience, or cold hardiness, which changes throughout the dormancy season. This has led to the development of predictive cold-hardiness models, which help farmers decide when to deploy expensive frost-mitigation measures. Unfortunately, cold-hardiness data for model training is only available for some fruit cultivars due to the need for specialized equipment and expertise. Rather, farmers often do have years of phenological data (e.g. date of budbreak) that they regularly collect for their crops. In this work, we introduce a new transfer-learning framework, Transfer via Auxiliary Labels (TAL), that allows farmers to leverage the phenological data to produce more accurate cold-hardiness predictions, even when no cold-hardiness data is available for their specific crop. The framework assumes a set of source tasks (cultivars) where each has associated primary labels (cold hardiness) and auxiliary labels (phenology). However, the target task (new cultivar) is assumed to only have the auxiliary labels. The goal of TAL is to predict primary labels for the target task via transfer from the source tasks. Surprisingly, despite the vast literature on transfer learning, to our knowledge, the TAL formulation has not been previously addressed. Thus, we propose several new TAL approaches based on model selection and averaging that can leverage recent deep multi-task models for cold-hardiness prediction. Our results on real-world cold-hardiness and phenological data for multiple grape cultivars demonstrate that TAL can leverage the phenological data to improve cold-hardiness predictions in the absence of cold-hardiness data.
20.
Ozmen Erkin Kokten; Raviv Raich
Maximum Likelihood Estimation of Stable ARX Models using Randomized Coordinate Descent Proceedings Article
In: ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 1–5, 2025, (ISSN: 2379-190X).
Abstract | Links | BibTeX | Tags: autoregressive model with exogenous variables, Convergence, coordinate descent, Data models, Finance, Numerical models, Numerical stability, optimization, parameter estimation, Signal processing, Soil Water Content, Speech processing, stability, Stability criteria
@inproceedings{kokten_maximum_2025,
title = {Maximum Likelihood Estimation of Stable ARX Models using Randomized Coordinate Descent},
author = {Ozmen Erkin Kokten and Raviv Raich},
url = {https://ieeexplore.ieee.org/abstract/document/10888613/authors},
doi = {10.1109/ICASSP49660.2025.10888613},
year = {2025},
date = {2025-04-01},
urldate = {2025-04-01},
booktitle = {ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
pages = {1\textendash5},
abstract = {Autoregressive models play an important role in a variety of applications including finance, engineering, sciences, and agriculture. While for some models (e.g., physics-based models) parameters are known, in other domains the parameters may not be available. This paper deals with the estimation of the parameters of an autoregressive model with exogenous variables. A significant body of literature has explored autoregressive model estimation across different estimation criteria, data availability, and parameterization; however, limited attention has been given to the estimation problem under stability constraints. The incorporation of stability constraints often results in increased computational complexity. As an efficient alternative, we propose to estimate stable ARX parameters using randomized coordinate descent. To demonstrate the efficiency of the proposed approach, we present an empirical convergence study and compare our approach to a state-of-the-art alternative.},
note = {ISSN: 2379-190X},
keywords = {autoregressive model with exogenous variables, Convergence, coordinate descent, Data models, Finance, Numerical models, Numerical stability, optimization, parameter estimation, Signal processing, Soil Water Content, Speech processing, stability, Stability criteria},
pubstate = {published},
tppubtype = {inproceedings}
}
Autoregressive models play an important role in a variety of applications including finance, engineering, sciences, and agriculture. While for some models (e.g., physics-based models) parameters are known, in other domains the parameters may not be available. This paper deals with the estimation of the parameters of an autoregressive model with exogenous variables. A significant body of literature has explored autoregressive model estimation across different estimation criteria, data availability, and parameterization; however, limited attention has been given to the estimation problem under stability constraints. The incorporation of stability constraints often results in increased computational complexity. As an efficient alternative, we propose to estimate stable ARX parameters using randomized coordinate descent. To demonstrate the efficiency of the proposed approach, we present an empirical convergence study and compare our approach to a state-of-the-art alternative.
