Post 2053599 by kjr@qoto.org
 (DIR) More posts by kjr@qoto.org
 (DIR) Post #2050485 by alan@mastodon.technology
       2018-12-17T08:32:48Z
       
       0 likes, 0 repeats
       
       The more I study and understand deep learning the less I like it
       
 (DIR) Post #2050486 by preslavrachev@mastodon.technology
       2018-12-17T10:13:32Z
       
       0 likes, 1 repeats
       
       @alan I also feel the same way. My issue is the interpretability of DL models. Especially, when developed by people who lack knowledge in some basic ML concepts. This has frustrated me enough that I started a movement around what I call Shallow Learning. Most of what people think they need a DL model for, could easily be a good fit for logistic regression or some simple decision tree logic.
       
 (DIR) Post #2053598 by alan@mastodon.technology
       2018-12-17T11:07:46Z
       
       0 likes, 0 repeats
       
       @preslavrachev I perfectly agree the only use I foresee for deep learning is computer vision and NLP to some extent (not everyone needs Word2Vec to get more than decent results)
       
 (DIR) Post #2053599 by kjr@qoto.org
       2018-12-17T14:34:21Z
       
       0 likes, 0 repeats
       
       @alanIn some áreas of NLP I'm not really impressed. I have see good results if sequencial models in Sentiment Detection, convolutional in Named Entity Extraction, or some DL approaches in text classification...  but with not a lit of effort I was able to improve them just with Logistic Regression (for sentiment), CRF for entities or SVM for classification. In other NLP areas like dialogue models or machine translation DL is offering better results. @preslavrachev