About

Kavita Bala is the 17th Provost of Cornell University, a position she assumed on January 1, 2025. She is a distinguished computer scientist, entrepreneur, and professor with a strong record of leadership and scholarship. Prior to becoming provost, Bala served as the inaugural dean of the Cornell Ann S. Bowers College of Computing and Information Science and as chair of Cornell's Department of Computer Science. During her tenure as dean, she secured the naming gift for the college, led a significant expansion of faculty to support rapid growth, and initiated construction of a new 135,000-square-foot building designed for robotics labs, experiential learning spaces, and faculty offices. She also played a key role as lead dean of the Cornell AI Initiative, advancing academic programs including new minors in AI and AI in Society, and helped establish the Schmidt AI in Science postdoctoral program. Additionally, she co-led a university-wide task force that developed guidelines for the responsible use of generative AI in education and learning. Bala's foundational research spans computer vision, computer graphics, and artificial intelligence, with a focus on image understanding, including the recognition of materials, styles, and object attributes; modeling of complex materials; and the use of crowdsourced training data. Her groundbreaking work on style recognition using deep learning led to the co-founding of GrokStyle, a visual recognition AI company acquired by Facebook in 2019. Her research contributions have been recognized by election to the American Academy of Arts & Sciences and by induction as a Fellow of the Association for Computing Machinery (ACM) and the SIGGRAPH Academy. She has received numerous awards including the SIGGRAPH Computer Graphics Achievement Award and the IIT Bombay Distinguished Alumnus Award, as well as multiple teaching awards at Cornell. Bala holds a Bachelor of Technology in Computer Science and Engineering from the Indian Institute of Technology, Bombay, and both a Master of Science and a Ph.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology. Her leadership and scholarship have positioned her as a national leader in computing and AI education and research.

Research topics

• Computer Science
• Artificial Intelligence
• Computer vision
• Algorithm
• Mathematics
• Medicine
• Computer graphics (images)

Selected publications

• MONITRS: Multimodal Observations of Natural Incidents Through Remote Sensing

  ArXiv.org · 2025-07-22

  preprintOpen access

  Natural disasters cause devastating damage to communities and infrastructure every year. Effective disaster response is hampered by the difficulty of accessing affected areas during and after events. Remote sensing has allowed us to monitor natural disasters in a remote way. More recently there have been advances in computer vision and deep learning that help automate satellite imagery analysis, However, they remain limited by their narrow focus on specific disaster types, reliance on manual expert interpretation, and lack of datasets with sufficient temporal granularity or natural language annotations for tracking disaster progression. We present MONITRS, a novel multimodal dataset of more than 10,000 FEMA disaster events with temporal satellite imagery and natural language annotations from news articles, accompanied by geotagged locations, and question-answer pairs. We demonstrate that fine-tuning existing MLLMs on our dataset yields significant performance improvements for disaster monitoring tasks, establishing a new benchmark for machine learning-assisted disaster response systems. Code can be found at: https://github.com/ShreelekhaR/MONITRS

  PublisherOA PDFDOI

• DiSciPLE: Learning Interpretable Programs for Scientific Visual Discovery

  2025-06-10

  article

  Visual data is used in numerous different scientific work-flows ranging from remote sensing to ecology. As the amount of observation data increases, the challenge is not just to make accurate predictions but also to understand the underlying mechanisms for those predictions. Good interpretation is important in scientific workflows, as it allows for better decision-making by providing insights into the data. This paper introduces an automatic way of obtaining such interpretable-by-design models, by learning programs that interleave neural networks. We propose DiSciPLE (Discovering Scientific Programs using LLMs and Evolution) an evolutionary algorithm that leverages common sense and prior knowledge of large language models (LLMs) to create Python programs explaining visual data. Additionally, we propose two improvements: a program critic and a program simplifier to improve our method further to synthesize good programs. On three different real-world problems, DiSciPLE learns state-of-the-art programs on novel tasks with no prior literature. For example, we can learn programs with 35% lower error than the closest non-interpretable baseline for population density estimation.

  PublisherDOI

• Analysis of ZVS Range of Single-Phase Shift Modulation for Bidirectional Dual Active Bridge DC-DC Converter in Electric Vehicle

  Lecture notes in electrical engineering · 2025-01-01

  book-chapter
  PublisherDOI

• DiSciPLE: Learning Interpretable Programs for Scientific Visual Discovery

  ArXiv.org · 2025-02-14

  preprintOpen access

  Visual data is used in numerous different scientific workflows ranging from remote sensing to ecology. As the amount of observation data increases, the challenge is not just to make accurate predictions but also to understand the underlying mechanisms for those predictions. Good interpretation is important in scientific workflows, as it allows for better decision-making by providing insights into the data. This paper introduces an automatic way of obtaining such interpretable-by-design models, by learning programs that interleave neural networks. We propose DiSciPLE (Discovering Scientific Programs using LLMs and Evolution) an evolutionary algorithm that leverages common sense and prior knowledge of large language models (LLMs) to create Python programs explaining visual data. Additionally, we propose two improvements: a program critic and a program simplifier to improve our method further to synthesize good programs. On three different real-world problems, DiSciPLE learns state-of-the-art programs on novel tasks with no prior literature. For example, we can learn programs with 35% lower error than the closest non-interpretable baseline for population density estimation.

  PublisherOA PDFDOI

• Effects of Digital Communication on Segmented and Super-Segmented Features of Pronunciation Patters among Undergraduate English Language Students in Nigeria

  Federal University Gusau Faculty of Education Journal · 2025-12-10

  articleOpen access1st authorCorresponding

  The widespread use of digital communication platforms, such as instant messaging, social media, and video conferencing, has transformed how university students interact and use language. This study investigates the effects of digital communication on pronunciation patterns, with a focus on both segmental features (vowel quality and consonant articulation) and supra segmental features (stress placement, rhythm, and intonation) among undergraduate students in Nigeria. A total of sixty participants, aged 18–25, and provided speech samples through structured reading tasks, spontaneous speech, and digital voice notes. Acoustic analyses measured vowel formants, consonant precision, stress placement, and pitch variation, while perceptual analyses involved listener ratings of intelligibility and naturalness. Results indicate that frequent engagement in digital communication is associated with reduced articulation precision, weakened stress contrasts, and a narrower pitch range, particularly in spontaneous and digital speech samples. These changes negatively impact intelligibility, suggesting that informal, rapid, and text-mediated communication encourages adaptations that deviate from standard pronunciation norms. Nevertheless, exposure to varied speech contexts can mitigate some comprehension challenges. The findings have important implications for language teaching, pronunciation training, and digital literacy programmes, highlighting the need to balance informal digital speech habits with strategies that preserve clear and intelligible communication. By understanding how digital communication influences pronunciation, educators and speech professionals can better support students in maintaining effective oral communication across both online and face-to-face contexts.

  PublisherOA PDFDOI

• Smart Model for Recognizing Hydroponic Lettuce Shortages Utilizing Higher Level Efficient Algorithms and Consensus Multi-Dilated Adaptive Networks

  2024-10-08

  articleSenior author

  The evolution that never stops, especially in smart agriculture, especially hydroponic farming has bought a great change for efficiencies and productivities. Nevertheless, identification and categorization of nutrient deficiencies in hydroponic lettuce are still challenging, specifically, when training deep learning models for low power ‘Edge’ IoT devices such as smart mobile phones and Raspberry Pi. Current methods are challenging due to high computational complexity, high power consumption, and inadequate utilization of resources, which hinders real-life implementation in embedded systems. To overcome these challenges, this paper presents an intelligent, highly efficient approach for the identification of nutrient deficiencies in hydroponic lettuce using advanced algorithms and a Consensus Multi-Dilated Adaptive Network (CMDA-Net). The proposed model harnesses or combines many of the current best-known or best performing approaches to yield a proper feature extraction and classification. This approach takes advantage of deep neural networks including VGG-16, Residual Attention Network (RAN) and Convolutional Neural Networks (CNNs) with a multi-dilated convolutional network to improve spatial feature learning while at the same time reducing computational cost. These features are then passed through LSTM network that help capture temporal information control and ensure that lettuce deficiencies are correctly classified. The entire system structure is enhanced through the new Improvised Honey Badger Algorithm (I-HBA), which adjusts specific networks’ parameters for maximum effectiveness with minimal resource consumption. Many experiments were carried out on Lettuce NPK dataset which contains test images of Fully Nutritional, Nitrogen Deficient, Phosphorus Deficient and Potassium Deficient lettuce. It respectively assessed the accuracy, specificity rate, precision, NPV, FPR, and MCC of the model. We also gained compelling improvements in terms of both computational time and model accuracy with exactly terminal prediction accuracy 96% and specificity 95%.

  PublisherDOI

• Scale-Aware Recognition in Satellite Images under Resource Constraints

  arXiv (Cornell University) · 2024-10-31

  preprintOpen accessSenior author

  Recognition of features in satellite imagery (forests, swimming pools, etc.) depends strongly on the spatial scale of the concept and therefore the resolution of the images. This poses two challenges: Which resolution is best suited for recognizing a given concept, and where and when should the costlier higher-resolution (HR) imagery be acquired? We present a novel scheme to address these challenges by introducing three components: (1) A technique to distill knowledge from models trained on HR imagery to recognition models that operate on imagery of lower resolution (LR), (2) a sampling strategy for HR imagery based on model disagreement, and (3) an LLM-based approach for inferring concept "scale". With these components we present a system to efficiently perform scale-aware recognition in satellite imagery, improving accuracy over single-scale inference while following budget constraints. Our novel approach offers up to a 26.3% improvement over entirely HR baselines, using 76.3% fewer HR images.

  PublisherOA PDFDOI

• Enhancing syphilis diagnosis through innovative adaptation of wet mount microscopy

  Indian Journal of Dermatology Venereology and Leprology · 2024-08-01 · 2 citations

  articleOpen access
  PublisherDOI

• AllClear: A Comprehensive Dataset and Benchmark for Cloud Removal in Satellite Imagery

  2024-01-01

  article1st authorCorresponding
  PublisherDOI

• AllClear: A Comprehensive Dataset and Benchmark for Cloud Removal in Satellite Imagery

  arXiv (Cornell University) · 2024-10-31 · 1 citations

  preprintOpen accessSenior author

  Clouds in satellite imagery pose a significant challenge for downstream applications. A major challenge in current cloud removal research is the absence of a comprehensive benchmark and a sufficiently large and diverse training dataset. To address this problem, we introduce the largest public dataset -- $\textit{AllClear}$ for cloud removal, featuring 23,742 globally distributed regions of interest (ROIs) with diverse land-use patterns, comprising 4 million images in total. Each ROI includes complete temporal captures from the year 2022, with (1) multi-spectral optical imagery from Sentinel-2 and Landsat 8/9, (2) synthetic aperture radar (SAR) imagery from Sentinel-1, and (3) auxiliary remote sensing products such as cloud masks and land cover maps. We validate the effectiveness of our dataset by benchmarking performance, demonstrating the scaling law -- the PSNR rises from $28.47$ to $33.87$ with $30\times$ more data, and conducting ablation studies on the temporal length and the importance of individual modalities. This dataset aims to provide comprehensive coverage of the Earth's surface and promote better cloud removal results.

  PublisherOA PDFDOI

Recent grants

• CGV: Medium: Collaborative Research: Understanding Translucency: Physics, Perception, and Computation

  NSF · $398k · 2012–2016

• CHS: Medium: Collaborative Research: Physics and Learning Integration Using differentiable rendering

  NSF · $400k · 2019–2023

• CPA -G&V: Collaborative Research: Visual Equivalence: a New Foundation for Perceptually-Based Rendering of Complex Scenes

  NSF · $275k · 2008–2012

• CHS: Small: Data-Driven Material Understanding and Decomposition

  NSF · $510k · 2016–2020

• CAREER: Scalable Rendering for Visual Realism in Scale-Complex Scenes

  NSF · $457k · 2007–2013

Frequent coauthors

• Bruce Walter

  Cornell University

  45 shared

• Shuang Zhao

  29 shared

• Noah Snavely

  Cornell University

  25 shared

• Hubert Lin

  Laboratoire d'Informatique de Paris-Nord

  24 shared

• Utkarsh Mall

  Columbia University

  20 shared

• Ganesh Ramanarayanan

  Cornell University

  19 shared

• Philip Dutré

  18 shared

• Bharath Hariharan

  17 shared

Labs

• Kavita Bala's LabPI

  Research in computer vision, computer graphics, and human perception

Education

• PhD, Computer Science

  Massachusetts Institute of Technology

  1999

Awards & honors

• Fellow of the American Academy of Arts & Sciences (2025)
• ACM Fellow (2019)
• Fellow of the SIGGRAPH Academy (2020)
• Computer Graphics Achievement Award (2020)
• NSF Faculty Early Career Development Award (CAREER)