LiVeAction: a Lightweight, Versatile, and Asymmetric Neural Codec Design for Real-time Operation
:::info Stub — Full Engineering Breakdown Coming This paper has a linked code implementation and was featured on Hugging Face Papers with 2 upvotes. A full breakdown with production viability rating, implementation notes, and honest limitations is being written. Subscribe to AI Letters → :::
| Authors | Dan Jacobellis & Neeraja J. Yadwadkar |
| Year | 2026 |
| HF Upvotes | 2 |
| arXiv | 2605.06628 |
| Download | |
| Code | https://github.com/ut-sysml/liveaction |
Abstract
Modern sensors generate rich, high-fidelity data, yet applications operating on wearable or remote sensing devices remain constrained by bandwidth and power budgets. Standardized codecs such as JPEG and MPEG achieve efficient trade-offs between bitrate and perceptual quality but are designed for human perception, limiting their applicability to machine-perception tasks and non-traditional modalities such as spatial audio arrays, hyperspectral images, and 3D medical images. General-purpose compression schemes based on scalar quantization or resolution reduction are broadly applicable but fail to exploit inherent signal redundancies, resulting in suboptimal rate-distortion performance. Recent generative neural codecs, or tokenizers, model complex signal dependencies but are often over-parameterized, data-hungry, and modality-specific, making them impractical for resource-constrained environments. We introduce a Lightweight, Versatile, and Asymmetric neural codec architecture (LiVeAction), that addresses these limitations through two key ideas. (1) To reduce the complexity of the encoder to meet the resource constraints of the execution environments, we impose an FFT-like structure and reduce the overall size and depth of the neural-network-based analysis transform. (2) To allow arbitrary signal modalities and simplify training, we replace adversarial and perceptual losses with a variance-based rate penalty. Our design produces codecs that deliver superior rate-distortion performance compared to state-of-the-art generative tokenizers, while remaining practical for deployment on low-power sensors. We release our code, experiments, and python library at https://github.com/UT-SysML/liveaction .
Engineering Breakdown
Plain English
This paper presents LiVeAction, a neural codec designed to compress sensor data (images, audio, 3D medical scans, hyperspectral data) efficiently on resource-constrained devices like wearables and edge sensors. The key innovation is an asymmetric architecture that uses a lightweight encoder on the device and a heavier decoder on the server, trading compression compute burden for bitrate savings while maintaining quality for both human and machine perception tasks.
Key Engineering Insight
By moving computational complexity to the decoder side rather than keeping it symmetric, the authors enable real-time compression on power/compute-limited edge devices without sacrificing the rate-distortion performance that recent neural codecs achieve. This asymmetry is critical for deployment—you can afford expensive decoding server-side but not expensive encoding on a battery-powered sensor.
Why It Matters for Engineers
Edge ML deployments today struggle with the bandwidth-computation tradeoff: standard codecs like JPEG waste bits because they don't understand what downstream ML models actually need, while recent neural codecs require too much compute to run on the device itself. LiVeAction solves this by enabling machine-aware compression that runs efficiently on resource-constrained hardware, directly reducing cloud bandwidth costs and latency for remote sensing, medical imaging, and IoT applications.
Research Context
Prior work split into two camps: standardized human-perception codecs (JPEG, MPEG) that are efficient but format-locked, and recent generative neural codecs that model signal dependencies well but are over-parameterized for edge deployment. LiVeAction advances the field by introducing a practical asymmetric design that bridges this gap—it exploits learned signal structure (like neural codecs) while remaining deployable on real constrained devices, enabling machine-perception optimization for non-traditional modalities that standard codecs never supported.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::