CV

PhD Research Intern

Dolby Advanced Technology Group (ATG)

Extended and improved predictive ML model maintenance.

• Developed a feature-based temperature scaling technique with a novel calibration loss that calibrates both probabilities and uncertainty under drift, improving prediction reliability.
• Expanded predictive maintenance by incorporating granular loss levels and adaptive loss bounds to optimize retraining decisions; explored extensions for concept drift.

Research Assistant

University of Massachusetts Amherst — DREAM Lab

Predictive machine learning (ML) model maintenance in production settings.

• Built an uncertainty-aware ensemble metamodel that forecasts deployed ML model performance without ground truth and tracks data drift’s direct impact on loss.
• Developed a predictive retraining orchestrator that retrains only when needed based on performance forecasts, minimizing downtime, reducing drift-induced errors, and lowering compute costs.

Technical Associate

Robert Bosch Center for Cyber Physical Systems & ArtPark, IISc (Aham Avatar XPRIZE Team)

Telepresence robotics R&D for the ANA Avatar XPRIZE.

• Led design and development of a multi-robot telepresence system controllable via a low-latency web interface.
• Built a motion-tracking framework to map human arm poses to dual 7-DoF robotic arms via inverse kinematics.
• Team achieved Semifinalist status in the $10M ANA Avatar XPRIZE.

MS/PhD

University of Massachusetts Amherst

Computer Science

B.E.

Ramaiah Institute of Technology

Computer Science and Engineering

• Probability
• Statistics
• Linear Algebra
• Machine Learning
• Deep Learning
• Artificial Intelligence
• High Performance Computing

Supervised ECG wave segmentation using convolutional LSTM

ICT Express

Malali, A., Hiriyannaiah, S., Siddesh, G. M., Srinivasa, K. G., & Sanjay, N. T. (2020). Supervised ECG wave segmentation using convolutional LSTM. ICT Express, 6(3), 166–169.

Scout: Predictive maintenance of machine learning models

In submission

Malali, A., Haas, P., Diao, Y., Mehra, A., & Maneriker, P. (2025). Scout: Predictive maintenance of machine learning models. In submission; manuscript available on request.

Perceptually guided zero-shot saliency for scalable visualization

Training-free saliency leveraging multi-layer activations from pretrained foundation vision models to approximate human perception and drive adaptive downsampling of large scatter plots, preserving salient structure while reducing render cost.

NeuralGLAZE: Noise generation for artwork using neural networks

Single-pass, perception-aware (LPIPS) noise generator that keeps images visually unchanged to humans while making artworks unusable to image classifiers—achieving modest performance at significantly lower computational cost compared to GLAZE.