Amar Phanishayee

Title: **Amar Phanishayee: Innovating Neural Network Technologies**

Introduction

Amar Phanishayee is a prominent inventor based in Seattle, WA, and a key figure in the realm of neural network technologies. With an impressive portfolio of 15 patents, he has made significant advancements in the field, focusing on enhancing the efficiency and effectiveness of neural network accelerators.

Latest Patents

Phanishayee's latest patents include groundbreaking work on "Neural network activation compression with non-uniform mantissas" and "Adjusting activation compression for neural network training." The first patent discloses an apparatus and methods for training a neural network accelerator using quantized precision data formats. It specifically addresses the storage of activation values from a neural network in a compressed format that may incorporate lossy or non-uniform mantissas, optimizing forward and backward propagation training.

In detail, the disclosed technology outlines a computing system with processors, memory, and a compressor, designed to perform forward propagation for a layer of a neural network. This system generates first activation values in a block floating-point format, which are then converted to a second block floating-point format with non-uniform and/or lossy mantissa for efficient storage in memory.

The second patent emphasizes methods for adjusting floating-point formats used during training. Here, the technology enables a computing system to select a second floating-point format based on a performance metric and convert activation values accordingly, aiming to enhance the performance metrics of neural networks.

Career Highlights

Amar Phanishayee is currently employed at Microsoft Technology Licensing, LLC, where he collaborates on innovative projects aimed at revolutionizing neural network processing. His extensive research and practical experience contribute to the development of state-of-the-art technologies that improve computational efficiency and data management in neural networks.

Collaborations

Throughout his career, Phanishayee has worked closely with notable coworkers, including Daniel Lo and Eric S. Chung. Together, they have fostered a collaborative work environment that encourages the exchange of ideas and accelerates the pace of innovation in their field.

Conclusion

In conclusion, Amar Phanishayee stands out as a visionary inventor in the domain of neural networks, with a strong focus on enhancing the capabilities of machine learning through innovative compression methods. His contributions at Microsoft Technology Licensing, LLC, and extensive patent portfolio exemplify his commitment to pushing the boundaries of technology and improving how neural networks are trained and utilized.