Amazing work.
— Jeremy Howard (@jeremyphoward) January 29, 2020
PS: we're all so screwed. https://t.co/lCIPJj3QE0
Amazing work.
— Jeremy Howard (@jeremyphoward) January 29, 2020
PS: we're all so screwed. https://t.co/lCIPJj3QE0
Fun fact: Thinc actually implements all of the practical ideas from my "Let Them Write Code" keynote!
— Ines Montani 〰️ (@_inesmontani) January 28, 2020
1. Callbacks
2. Function registries
3. Entry points
4. Single-dispatch
🔮 Docs: https://t.co/8wfmqQQMhH
🖼 Slides: https://t.co/yrYbChgrH4
📺 Video: https://t.co/ddQOBQB7o9 https://t.co/3d6zNBBzXo
Even my continued fiddling with the SHA-RNN model shows there's a _lot_ to be studied and explored. I haven't published new incremental progress but you can tie the RNN across the 4 layers to substantially decrease total params yet get nearly equivalent perplexity results.
— Smerity (@Smerity) January 28, 2020
Tomorrow morning we'll be back to working on our Dialect Quiz bot. No spoilers, but I've been working on it a bit on my own and I've... some good news and some bad news. 👀😳
— Rachael Tatman (@rctatman) January 28, 2020
Come hang out with me tomorrow! 💻🤓☕️https://t.co/RM622mzSGQ
Open-domain conversation is an extremely difficult task for ML systems. Meena is a research effort at @GoogleAI in this area. It's challenging, but we are making progress towards more fluent and sensible conversations.
— Jeff Dean (@JeffDean) January 28, 2020
Nice work, Daniel, @lmthang & everyone involved! https://t.co/nqpsBKlpkl pic.twitter.com/vjkXoSUHf3
I feel like good-old LSTM (or QRNN) are usually better for text classification indeed.
— Thomas Wolf (@Thom_Wolf) January 28, 2020
Note that for those who want to give a try at text classification with pretrained Bert models, you can give a look at the experimental section of this paper https://t.co/w4MWPTB79u
The other dimension here is monolingual vs multilingual models. I think monolingual models in low-resource languages currently have an edge e.g. as seen in our MultiFiT paper.
— Sebastian Ruder (@seb_ruder) January 28, 2020
A lot has been said about the dangers of large-scale language model trained on the internet, like GPT-2.
— Mark O. Riedl (@mark_riedl) January 27, 2020
We fine-tuned a version of GPT-2 to avoid generating descriptions of non-normative behavior (killing, suicide, things inconsistent with social norms) https://t.co/V12NM8HSc7 pic.twitter.com/5GCMtjnEM6
Nice job by @GaryMarcus https://t.co/v7VliOtzbT. I agree on every point.
— Thomas G. Dietterich (@tdietterich) January 25, 2020
(Well, except for the one reference to "genuine understanding"...) 1/
Plato Dialogue System – a Flexible Conversational AI Research Platform
— ML Review (@ml_review) January 25, 2020
By @Uber
Bridges gap between research & production.
Supports text/speech, jointly-trained components, single- or multi-party interactions, offline/online training
GitHub: https://t.co/FMnhIVcmCc pic.twitter.com/PZU0suDDTn
Hey machine learning enthusiasts! Want to read about how we use neural networks to recommend easy issues to new OSS contributors? Sure you do! 😃 Check out our latest blog post here https://t.co/JkOP6Fhrp5 pic.twitter.com/ektYXKerZ8
— GitHub (@github) January 22, 2020
Reformer: The Efficient Transformer
— Thomas Lahore (@evolvingstuff) January 20, 2020
"we replace dot-product attention by one that uses locality-sensitive hashing, changing its complexity from O(L^2) to O(L log L), where L is the length of the sequence"
paper: https://t.co/3o1scnoCCT
code: https://t.co/OjLbTyILln