50+ days of Machine Learning

 Hello :) Today is 130! A quick summary of today: covered Lecture 22 : MAP, Lecture 23 : Naive Bayes, Lecture 24 : Logistic Regression from Stanford's CS109 From 11pm later, to 3am (I believe) is the 3rd day of MLx Fundamentals , and I will share about the 2 sessions tomorrow.  Lecture 22: MAP We saw new method for optimization: gradient descent (or ascent for now). We want to choose theta, that is the argmax of some function. Seeing the blue curve is nice, but what can we do if we cannot see it ? Imagine we start with a random theta value, we look at the likelihood of that theta, and we can know the derivative at that point, and based on that we can follow along where the derivative is positive. When we reach the top, we actually don’t know we are at the top but when we check the gradient is 0. So we assume we are at the top. In the real world, because we want to think of minimising the loss, we minimise the negative log likelihood. MLE is great, but it has a problem. It likes to

 Hello :) Today is Day 129! A quick summary of today: I thought to myself: is there an AI that can speak with Scottish accent and does not sound robotic covered Lecture 21  from Stanford's CS109 on Maximum Likelihood Estimation Firstly, about the Scottish AI I was looking for videos to practice my Scottish accent (which is awful). After a few, I thought whether there has been an AI model that given text, can generate that text being said in Scottish accent. Started looking, and the best I could find was this , which tries to do it but it maybe sounds 5% (general) Scottish.  Then, I started looking on huggingface and kaggle, for models developed by solo devs - nothing. I searched TTS models, most are trained on general south UK English, American, Indian, Australian. But none can speak in Scottish.  Then, I started looking for a dataset. I could not find any either. That is when I tried something. I found one video on youtube where there is only Scottish accent used - The Glasgow Uni