Top all-time · #7273
Silicon Valley AI Lab by svail
Company blog · svail.github.ioaisystems
Silicon Valley AI Lab by svail is the company blog at svail.github.io, writing about ai and systems. Since 2015, 11 different Hacker News readers have linked it 12 times, most often “Getting Deep Speech to Work in Mandarin”. No feed was found, so read it on the site itself.
Too few sampled comments (4) to say how often HN recommends it.?
Why HN links it
Links to posts from Baidu's Silicon Valley AI Lab on deep learning performance, including GPU batch-size and RNN benchmarks, DeepBench, and the Deep Voice text-to-speech work.
Summarised by Claude from a sample of 4 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- Getting Deep Speech to Work in Mandarin3 people · 3 mentions · first linked 2016-02-10
- Baidu Research Announces New Open Source Deep Learning Benchmark3 people · 3 mentions · first linked 2016-09-26
- Persistent RNNs3 people · 3 mentions · first linked 2016-03-28
What Hacker News says?
That all sounds reasonable. I was surprised at how much the batch size can matter on different hardware. Maybe you've seen this post: http://svail.github.io/rnn_perf/ It's primarily RNN-focused, but the discussion about batch sizes on GPUs is interesting.
Comment linking /rnn_perf/
on “Leaf: Machine learning framework in Rust”Submitted Getting Deep Speech to Work in Mandarin as a story · 58 points
HN discussionThank you! For fully end-to-end models, it's hard to say exactly. The Char2Wav paper demonstrates that there is hypothetically an architecture and a set of weights that can do synthesis end-to-end, but we cannot yet train such a system. On Reddit, one of the Char2Wav authors comments that they tried training it directly and didn't get great results, and at SVAIL we've also had some trouble doing so. I think it is ve…
Comment linking /persistent_rnns/
on “Deep Voice: Real-Time Neural Text-To-Speech”Submitted Optimizing Recurrent Neural Network Performance as a story · 21 points
HN discussionBaidu Research also recently announced DeepBench: https://svail.github.io/DeepBench/
Comment linking /DeepBench/
on “Deep Learning GPU Benchmarks: GTX 1080 vs. Titan X Pascal”Submitted Recognizing Both English and Mandarin with a Single Algorithm as a story · 16 points
HN discussionI'm not exactly sure what you mean, but it sounds like sarcasm. 1) It isn't actually obvious that if you can train a model faster this will allow you to get better final accuracy. It obviously allows you to get to a result faster, but only by being able to do many of these experiments and learning from them are you able to translate faster training into better accuracy. 2) The research that allows the training to be…
Comment linking /persistent_rnns/
on “Baidu Researcher Pushes GPU Scalability for Deep Learning”Submitted Baidu Research Announces New Open Source Deep Learning Benchmark as a story · 5 points
HN discussionSubmitted Efficient Neural Networks Without the Math as a story · 4 points
HN discussionSubmitted DeepBench – deep learning hardware benchmark as a story · 3 points
HN discussionSubmitted Persistent RNNs as a story · 2 points
HN discussionSubmitted Around the World in 60 Days: Getting Deep Speech to Work in Mandarin as a story · 1 points
HN discussion
