Deep learning job postings have collapsed in the past six months

Re:Deep learning job postings have collapsed in the past six... by bratao on Monday August 31, 2020 @10:08AM (#24331579)

I do not agree with this. I work heavily with NLP models for production in the Legal domain (where my baseline is where a 8GB 1080 must predict more than 1000 words/sec). This year was when our team glued enough pieces of Deep Learning to outperform our previous statistic/old ML pipeline that was been optimized for years.

Little things compound such as optimizers ( Ranger/Adahessian), better RNN ( IndRNN, Linear Transformers, Hopfield networks ) and techniques (cache everywhere, Torch script,gradient accumulation training)

[ Parent | on HN ]

Re:Deep learning job postings have collapsed in the past six... by danieldk on Monday August 31, 2020 @01:47PM (#24333990)

I do not agree with this. I work heavily with NLP models for production in the Legal domain (where my baseline is where a 8GB 1080 must predict more than 1000 words/sec).

What kind of network are you using? I can do near-SoTA multi-task syntax annotation [1] with ~4000 tokens/s (~225 sentences/s) on a CPU with 4 threads using a transformer. Predicting 1000 words/second on a reasonably modern a GPU is easy, even with a relatively deep transformer network.

[1] 8 tasks, including dependency parsing.

[ Parent | on HN ]

Re:Deep learning job postings have collapsed in the past six... by maxlamb on Monday August 31, 2020 @11:12AM (#24332268)

Interesting. What's the main goal(s) of your NLP models?

[ Parent | on HN ]

Re:Deep learning job postings have collapsed in the past six... by bratao on Monday August 31, 2020 @11:41AM (#24332564)

We work on multiple models, all related to legal proceedings and lawsuits, such as: - Structure Judicial Federal Register texts - Identify entities in Legal texts (citation to laws, other lawsuits) - Predict time to completion, risk and amount due of a lawsuit - Classifying judicial proceedings to non lawyers

[ Parent | on HN ]

Re:Deep learning job postings have collapsed in the past six... by grumple on Monday August 31, 2020 @02:27PM (#24334423)

How accurate is your prediction of time/risk/amount? How useful is identifying entities or classifying proceedings?

[ Parent | on HN ]