3 ms·
Reminds me of positional encoding in attention based networks.
by anon291 4y ago
Reminds me of positional encoding in attention based networks.
- johntb86 4y agoI was thinking that too, but this description makes it seem more like a type of windowing rather than the position encoding in a transformer (which is fixed): "the information [...] gets passed between different neural populations in a predictable way, which serves to time-stamp each sound with its relative order.”