Paras Jain

A patent outlining systems and methods for training machine models using augmented data.

Accelerates wide-area transfers in the cloud by 5x via overlay routing and parallelism, allowing reliable and efficient movement of terabytes of data at lower costs.

A patent focusing on optimizing neural network structures for embedded systems.

Enables training large neural networks like BERT on memory-scarce battery-operated edge devices.

Improves the power-efficiency of TPUs by 4-6% by synthesizing a novel low-power approximate version using learning-augmented search.

Focuses on learning to represent software functionality for automated software engineering tasks and improving the robustness of ML4Code.

Proposes a method to improve compositional generalization of language representations by grounding structured predictions with an attention mechanism.

Demonstrates that reinforcement learning can significantly accelerate first-order optimization, outperforming state-of-the-art solvers by up to 3x.

Explores how Transformers enable Graph Neural Networks to achieve state-of-the-art performance on graph classification tasks by representing long-range context.

Focuses on synthesizing low-power approximate inference accelerators using large-scale search techniques.

A patent describing multi-channel sensor simulation for autonomous control systems.

A method to use up to 5x less memory when training Deep Neural Networks by recomputing activations.

A patent related to data synthesis methods for autonomous control systems.

Demonstrates 2.5x-4.9x speedups for deep learning inference workloads through GPU multitenancy.

Achieves performance portability for hand-vectorized programs with up to a 1.88x speedup.

Demonstrates 2.5x-4.9x speedups for deep learning inference.