Implementing Semantic Communication (Lossy vs. Lossless)

Determine how to implement semantic communication—communication schemes that convey the meaning of data by computing semantics and transmitting shorter representations—and ascertain whether such implementations can be realized in lossy or lossless settings.

Background

In discussing alternatives to the Shannon paradigm, the paper references proposals for "semantic communication," where the idea is to transmit the meaning of data or compute semantics and send shorter representations. The authors note that despite the conceptual appeal, the concrete means of realizing such systems has not been established.

They further observe that attempts at semantic communication often rely on computable semantic functions, which they argue falls into a classical trap. This contrasts with their proposed Kolmogorov paradigm that leverages large models approximating the (uncomputable) Solomonoff prior.

References

While it is unclear how such a proposal would be implemented, lossy or lossless, the central idea is clear: we communicate the meaning of the data, or we compute the meaning of the data then send their shorter representations.

Lossless data compression by large models (2407.07723 - Li et al., 24 Jun 2024) in Background, Subsection "Solomonoff Prior"