Prasanna Sattigeri

Research Staff Member

I am a researcher at IBM Research AI and my broad research interests include machine learning, deep learning and signal processing. My current work focuses on exploring structure in the data using generative models and developing algorithms that are data efficient. My research also involves developing theory and practical systems for impactful machine learning applications that demand constraints such as robustness, fairness and interpretability.

Top Work

New tricks from old dogs: multi-source transfer learning

New tricks from old dogs: multi-source transfer learning

Transfer Learning

Publications with the MIT-IBM Watson AI Lab

Who Should Predict? Exact Algorithms For Learning to Defer to Humans
Who Should Predict? Exact Algorithms For Learning to Defer to Humans
 
Post-hoc Uncertainty Learning using a Dirichlet Meta-Model
Post-hoc Uncertainty Learning using a Dirichlet Meta-Model
 
Fair Infinitesimal Jackknife: Mitigating the Influence of Biased Training Data Points Without Refitting
Fair Infinitesimal Jackknife: Mitigating the Influence of Biased Training Data Points Without Refitting
 
Selective Regression under Fairness Criteria
Selective Regression under Fairness Criteria
 
Detector-Free Weakly Supervised Grounding by Separation
Detector-Free Weakly Supervised Grounding by Separation
 
StarNet: towards weakly supervised few-shot detection and explainable few-shot classification
StarNet: towards weakly supervised few-shot detection and explainable few-shot classification
 
Understanding Behavior of Clinical Models under Domain Shifts
Understanding Behavior of Clinical Models under Domain Shifts
 
Building Calibrated Deep Models via Uncertainty Matching with Auxiliary Interval Predictors
Building Calibrated Deep Models via Uncertainty Matching with Auxiliary Interval Predictors
 
Fair Selective Classification Via Sufficiency
Fair Selective Classification Via Sufficiency
 
AdaFuse: Adaptive Temporal Fusion Network for Efficient Action Recognition
AdaFuse: Adaptive Temporal Fusion Network for Efficient Action Recognition
 
OnlineAugment: Online Data Augmentation with Less Domain Knowledge
OnlineAugment: Online Data Augmentation with Less Domain Knowledge
 
TAFSSL: Task-Adaptive Feature Sub-Space Learning for few-shot classification
TAFSSL: Task-Adaptive Feature Sub-Space Learning for few-shot classification
 
AR-Net: Adaptive Frame Resolution for Efficient Action Recognition
AR-Net: Adaptive Frame Resolution for Efficient Action Recognition
 
Learning New Tricks From Old Dogs: Multi-Source Transfer Learning From Pre-Trained Networks
Learning New Tricks From Old Dogs: Multi-Source Transfer Learning From Pre-Trained Networks
 
New tricks from old dogs: multi-source transfer learning
New tricks from old dogs: multi-source transfer learning
 
Co-regularized Alignment for Unsupervised Domain Adaptation
Co-regularized Alignment for Unsupervised Domain Adaptation