Home

Dissertation Defense

Date: Friday, September 18th, 2026

Time: 1:00 pm - 3:00 pm CT

Location: EER 0.806/0.808 Mulva MPR. Public parking available in the San Jacinto Garage and the Speedway Garage

Zoom meeting link: https://utexas.zoom.us/j/84885961880

Title: Machine-Oriented Compression

Advisor: Dr. Neeraja J. Yadwadkar

Abstract:

Modern sensors—ranging from wearables and smartphones to medical devices and satellites—produce increasingly rich streams of high-resolution data. Existing media compression standards, or “codecs,” such as JPEG or MPEG can reduce storage and transmission requirements by orders of magnitude by discarding imperceptible details and eliminating redundancies.

However, these standards were designed under assumptions that no longer hold: (1) that a human, rather than a machine learning (ML) system, is the ultimate consumer of the signal; (2) that the signal must be fully decoded before any other system can use it; (3) that the signals of interest are limited to conventional media formats such as stereo audio and RGB images or video; (4) that encoding happens where power is plentiful and decoding where it is scarce; and (5) that applications can abide years—or decades—of standardization before benefiting from compression.

Under these assumptions, the vast majority of applications are faced with a difficult choice: (A) design and maintain a bespoke compression system, (B) use or repurpose an existing but ill-suited standard, or (C) rely exclusively on general-purpose, bit-reversible compression methods. Each option is prohibitive: (A) in engineering cost, (B) in bits squandered on irrelevant details while task-relevant information is discarded, and (C) in signals discarded entirely for want of a method that meets the application’s compute, power, bandwidth, and storage constraints.

To escape this trilemma, this dissertation proposes a design philosophy, attendant principles, and techniques for next-generation compression systems, which we refer to as machine-oriented compression. Machine-oriented compression is characterized by a design goal answering each of these assumptions: (1) preservation of features salient to machine perception, interoperating with unmodified pre-trained foundation models; (2) model training and inference directly on compressed representations, without additional decoding overhead; (3) end-to-end optimization and adaptation for specific sensors, signals, environments, and downstream tasks; (4) lightweight encoding suitable to run on low power and programmable NPUs, CPUs, or even MCUs; and (5) accessible, data-driven design of bespoke codecs with limited resources, while remaining compatible, when appropriate, with existing codecs and the decades of hardware and software built upon them.

In this defense, I present the first machine-oriented compression systems, instantiated across standard audiovisual media as well as signals ill-served by existing codecs, including spatial audio arrays, hyperspectral remote sensing data, and 3D medical images. I show how training and inference directly on compressed representations provides an efficiency multiplier when scaling the resolution of machine perception systems; how encoding-efficient asymmetric autoencoders enable variable-rate and progressive compression with encoders light enough for MCUs; how machine-oriented compression extends to real-time systems such as video-based autonomous robots, whose latency and reliability constraints have long precluded cloud-based processing; and how negative-distortion transcoding keeps these systems fully compatible with the entrenched hardware and software of existing standards. Together, these results demonstrate that compression can enable profoundly new capabilities for sensing and perception under extreme power, bandwidth, latency, or accuracy constraints.