Please enable JavaScript.

Coggle requires JavaScript to display documents.

Courses, control theory W/M I only need a fundamental, not expertise…

- - - - Deep learning -M/W
        (half will be first 1/3 of class is review,
        rest is new and not things that are learned easy)Deep neural networks. This class is like an overview of the foundational pieces of deep learning. I have have taken an abbreviated version of this class before but this would be more in depth. Deep learning is the AI you will see in the news (ability of a ML tool or algorithm to form an actual understanding of a task (not just spiting out probabilities like in stat learning) the what are the tools of deep learning
        - Convolution and recurrent NNs: Most original image recognition tools. the bread butter of NN
        -Transformers Heavily used in natural language processing, NLP (computers ability to understand language) ex. applications include design things like ChatGPT / other common NLP algos. The process of predicting the next word that will be said/typed etc.
        -Encoder/decoder used to create latent spaces (taking a large amount of data and representing with only a few data points). Use to teach deep learning algos about how similar to objects are. My capstone (current research experience) actually involved a pretty simple application of an encoder.
        -Generative networks used for image / text generation. example applications include being given a piece of text and then creating the image.
        -GANs adversarial networks, a practice of training 2 models, one of which is "judging" the other. This is commonly applied to improve the performance of different networks.
        LSTM helpful for getting a deep learning algos to understand both the large and small view (understand both what is the mostly like next word but also what is the theme of a sentence.
- - - - Linear Models -M/W
        still have seen this in SOME
        capacity before but I would say
        much deeper dive into things
        I have seen before). Always helpful to know but imo this is not something where the knowledge gained would justify a PhD
        
        Research
        most of the 1/2 half of the program will be spent doing research. The important thing here is that I want to be applying what I am learning in the upper level DE / ML courses to actual problems. research must be applied!! this is the type of work/research that will transfer much better to industry jobs.
        Worth noting the research itself is a skill / something taught as an actual class. From looking on online message boards where people have asked the question "do a i need a PhD to do deep learning", something people often respond with is that companies prefer the PhDs since they have actually been taught / have experience being given a problem and then having to systematically find the best solution using very technical tools.
        More understanding of this needed!
- - - - More advanced ML tools require more
        advanced infrastructure to support them. The type of models I run today (statistical learning models) do not require as complex of infrastructure. RL and deep learning does and, in addition to just learning the deep learning aspects, I should have a reasonable understanding of these topics.
        
        RL / NLP / Bayesian -W
        (new and very much type of work
        I would want to be doing long term)
        This course of much more focused on taking the previous course where you learning "what are tools for deep learning" and then move into "how to apply the tools".
        -RL (reinforcement learning) this is much more advanced than general statistical learning where you plug data into a algo, it learns the patterns and then spits out a prediction. RL is the process of AI being in some environment (ex. a car on the road) and then based on the feedback from how that AI interacts with the environment (ex. successfully driving down the street or crashing into something). not that much of an over simplification to say that this is teaching AI to learn the same way a human does, by experience.
        -Bayesian the idea of including prior knowledge or intuition in AI thought processes. Can prevent over reacting to a single event.
        NLP natural language processes, the ability of AI to understand text / language. (think a chatGPT tool). Personal opinion is that while tools like chatGPT are very powerful, this is a space with an astronomical amount of hype/misinformation at the moment. this will continue to be a place with a lot of promising applications to solving problems but will also have a lot of headline grabbing applications that don't actually work long term.