3 ms·
So would something like tensorflow-compress[1] v4 (at some point) be the ideal candidate for breaking the current record? [1] https://github.com/byronknoll/ten
by jonbaer 6y ago
So would something like tensorflow-compress[1] v4 (at some point) be the ideal candidate for breaking the current record?
[1] https://github.com/byronknoll/tensorflow-compress https://github.com/byronknoll/tensorflow-compress
- lifthrasiir 6y agoFor now GPU is prohibited, as high performance GPGPU capable of ML is not that widespread.
- svantana 6y agoThe current leader uses neural networks (LSTMs I believe) so that's obviously a good strategy. But since the size of the decoder gets counted, vanilla tensorflow is probably a bad choice.
- beaconstudios 6y agoyou can submit your compressor compressed too, if tensorflow networks lend themselves to compression.