>>13440065You hardly need any statistics to get started, but a good understanding of probabilities will help you understand a lot of it. My experience is that you can pick up most of it along the way. The big "new paradigm" is turning out to be transformers, which is why I linked that Yannic video in particular, but transformers are very hard to grasp intuitively so they're not a great place to start.
>>13440091ML is the domain of Indians implementing linear regression and random forests. Anyone can come in and do whatever because as long as they call it ML someone will pay for it. But I agree that the future is in a more signal-processing based model. Every time we find a way to pass information more clearly through networks, we set new benchmarks. Still waiting for someone to disentangle the signals in transformers.
