maziarraissi/applied-deep-learning

Applied Deep Learning Course

applied-deep-learning's Issues

Hey, just wondering if there were assignments for this. Or a syllabus/ overview.

Hey, just wondering if there were assignments for this. Or a syllabus/ overview.

Would you recommend a self-learner watch all the videos to learn or some other method?

EfficentNet Lecture Video Question

Hello Dr. Maziar Raissi, I am a PhD student in Machine Learning and sincerely appreciate you posting these videos online as they are very helpful. I have a small question that is not explained well in the paper and the authors skip over it. In the mnasnet paper, they use an empirical observation that doubling the Latency will increase accuracy by 5%: " for instance, we empirically observed doubling the latency usually brings about 5% relative accuracy gain." and therefore the authors solve for the omega parameter (setting alpha equal to beta) and obtain -0.07 from this empirical observation. However, this observation is only for LATENCY and not for flops or memory as the parameter they were optimizing was LATENCY. In the efficientnet paper the authors describe that they optimized flops instead of latency with an objective function (reward) as the following formula: ACC(m)*[FLOPS(m)/T]^w . With this reward function they use the same value for omega for controlling the trade off between accuracy and flops: "Specifically, we use the same search space as (Tan et al., 2019), and use ACC(m)×[FLOP S(m)/T]^w as the optimization goal, where ACC(m) and F LOPS(m) denote the accuracy and FLOPS of model m, T is the target FLOPS and w=-0.07 is a hyperparameter for controlling the trade-off between accuracy and FLOPS. Here, they use w=-0.07 which is based on the empirical observation from the mnasnet paper that doubling the LATENCY will increase accuracy by 5%. This observation is only for latency and not flops though. So my question is why they can make this assumption that w=-0.07 for flops as well unless this is some sort of guess for the w hyperparameter? Thanks, let me know if you would like me to elaborate on my confusion.

How I can start in

I have no idea in programming

Newer Models

Hello Dr. Raissi,

I really like your videos! Wondering if you will be reviewing any newer papers/models from NLP side of things for example GPT-4, PaLM, or LLama2?

Thanks,

Karl Gardner

Deep learning

Missing Lecture Videos

Thank you for this incredible resource!
Will the missing lecture videos (for example the ones under advanced topics in part 1) ever be posted?

Gh

Source code of these wonderful slides

Thanks a lot for sharing these awesome slides. I was surprised that in one page how nicely you put everything in an organized manner. Is it possible to share the source code of the slides? Or at Atleast one so that we can also make something similar.