Time permitting (and if there are some well-liked questions here) I will do an “AMA” (Ask Me Anything) in tonight’s class. So if you have some question (on any topic you like) that’s you’d like me to answer, here’s your chance - just reply on this thread! To save on answering questions I’ve already answered, please first take a look at my Reddit AMA - and if someone asks a question I answered there, you can help them out by posting a link to my answer.
I’ll answer questions that have lots of "like"s - so please click the “heart” if you see a question you want answered.
What are the up and coming DL/ML things (algorithms, techniques, papers, etc) that you are most excited about? You’ve mentioned last year that you were not a believer in reinforcement learning, do you still feel the same way?
Could a model today recreate an individual’s voice after a few recordings of learning? My thought behind this is that when people get diagnosed with certain diseases that take their voice away from them, it would be great to be able to help them talk with their own voice still through a computer program. Just wondering if the advancements that are needed in this field are already here or if we are still needing some improvements for this to happen.
Jeremy, have you changed your views on Reinforcement Learning (Deep RL of course) practical potential in the last few months? if yes, what research/papers got your attention in this field?
You and the team have put a tremendous amount of work and time into the videos, the fastai library and support and provide it all for free in an open way (and thank you, we all appreciate it!).
Many (if not all) of the popular ML algorithms used today are improved versions of algorithms invented decades ago. Do you think it is still possible for a new kind of ML algorithm to be invented? And I mean all tree-based algorithms (e.g., boosting trees) as one kind, all NN based algorithms as another kind, etc.
It is commonly perceived that elite consulting firms, like McKinsey, provide a platform where talented and ambitious young people well-rounded in both business and analytics get to solve a diverse set of exciting real-world challenges. You worked there for a long time and now seem to be rather disparaging about it.
What made you leave the management consulting field? It is on many people’s career list and it would be great to hear your view on it. For example, would you still recommend starting a data science career in a consulting firm?
One problem we have at work right now is putting models into production. Putting one model into production is not super difficult (not easy, but can be handled by brute force), but getting to a point where you have 10+ models in production is quite overwhelming to handle. Do you have any tips on managing a large number of models (tracking them, putting them into a system, building a review process around them)?
I know you had the change to assist last week to NeurIPS. Would you mind share your personal choices on interesting ideas/papers that were shared in the conference?
Now that MLPerf is out, do you think fastai will try their hand at the open division? (allows anything between dataset and trained result - so any new algorithm ideas are allowed).
What will replace Deep Learning? Deep learning could have not been adopted until we got large amounts of data and large computing power. Is there a technology now that may benefit from developments in the future? Loved the class! Can’t wait for second part.
With neural nets staring to perform better and better with general tabular data as well, should we still focus on traditional ML algorithms? (apart from Random Forest and Decision Trees which are helpful for determining feature importance)
If yes, then which ones?
Great course, thanks Jeremy! My question is: the second part of the course will be available as this one? I mean will be open to the public from differents parts of the world?
fastai DL courses related: for someone planning to take part 2 in 2019, what would you recommend doing/learning/practicing until the part 2 course starts?
To save on answering questions I’ve already answered, please first take a look at my Reddit AMA - and if someone asks a question I answered there, you can help them out by posting a link to my answer.