Pedro Domingos Says Cross-Entropy Is Log-Likelihood Not Compression
University of Washington professor replies to @ChrSzegedy on objective functions.
Pedro Domingos, Professor Emeritus of Computer Science at the University of Washington and author of The Master Algorithm, posted a reply directed at @ChrSzegedy. The message states that cross-entropy is log-likelihood, not compression. The exchange concerns objective functions. The packet records this as a reply from the academic account and includes his professional background in the source line. No other posts, replies, or outcomes appear in the supplied lines. The statement stands as the content Domingos posted rather than an established technical fact beyond the author's assertion.
@ChrSzegedy Cross-entropy is is log-likelihood, not compression.
Combined views
121
Pedro Domingos Says Cross-Entropy Is Log-Likelihood Not Compression
University of Washington professor replies to @ChrSzegedy on objective functions.
Pedro Domingos, Professor Emeritus of Computer Science at the University of Washington and author of The Master Algorithm, posted a reply directed at @ChrSzegedy. The message states that cross-entropy is log-likelihood, not compression. The exchange concerns objective functions. The packet records this as a reply from the academic account and includes his professional background in the source line. No other posts, replies, or outcomes appear in the supplied lines. The statement stands as the content Domingos posted rather than an established technical fact beyond the author's assertion.
@ChrSzegedy Cross-entropy is is log-likelihood, not compression.
Reactions
1 postCS 
Christian Szegedy@ChrSzegedy@pmddomingos It is also the compression rate when one uses the LM as a compressor (with aruthmetic coding).
- likes: 1
- replies: 0
- bookmarks: 0
- reposts: 0