I have done tons of machine learning projects. However, in quite traditional algorithms with some innovation of framework. I wanna invest some time to learn DL, but by looking at some JD, I feel there is lots of position claiming they need DL background, but doesn’t sound like they have the domian question needed to be solved by DL, RL, etc. Plus, based on the estimation of their data source, I doubt it would yield good results. I could be wrong, so wanna get your thougts on DL usage in industry.
I use it for finding hot girls on the Internet. I scrape photos and use a deep learning algo to rank girls and then send me the handles of the hottest ones so I can add it to my fap bank
The real killer app is to add a logistic regression to the pipeline that predicts the probability of you getting a reply if you dm. Though really it'll just be a one liner return 0
Nvidia’s AI podcast interviewed a guy why published a paper on optimizing for getting a reply on tinder. The guy put himself out there to collect an initial dataset of ground-truth a and train an initial estimator, then kept fine tuning it by using it
Tech Industry
Yesterday
840
Google vs Meta vs Apple vs Amazon
Tech Industry
2d
44466
Worried that our top performer is an attrition risk. How do managers handle this?
Tech Industry
Yesterday
1183
I haven’t done shit today!
Tech Industry
Yesterday
3304
Avoid teams with only Chinese or Indians especially with a Chinese/Indian manager
I use it for computer vision and it performs very well. However in other business cases traditional ml models give you more insight into how the features are being weighted which is an important signal to the business. For example if trying to retarget customers, even if a dl method gets slightly better results, a logistic regression coupled with lasso can inform the business that x features are the most important, leading them to invest in optimizing those features
This makes sense. Just for curiosity, I think I would still invest time to learn and leverage. Thanks!
Cv is the main use case for dl since the features are nonsense anyways (numpy array)