<ul><li>Implicit Neural Representations (INRs) are versatile for representing discretized signals and offer benefits such as infinite query resolution and reduced storage requirements.</li><li>A new compression algorithm called SINR is introduced, which uses high-dimensional sparse code within a dictionary to compress the vector spaces formed by weights of INRs.</li><li>The atoms of the dictionary used to generate the sparse code do not need to be learned or transmitted, resulting in substantial reductions in storage requirements for INRs.</li><li>SINR outperforms conventional INR-based compression techniques and maintains high-quality decoding across various data modalities.</li></ul>

SINR: Sparsity Driven Compressed Implicit Neural Representations

Discover more