Machine learning systems tune their internal parameters to achieve lower prediction errors through iterative adjustments.
The graph shows how cost varies with one weight parameter along a characteristic U-shaped curve whose bottom represents the minimum achievable cost. The black dot locates the starting weight, from which the dashed gradient line indicates the direction of steepest cost rise. Multiple arrows trace the incremental updates that move opposite the gradient, sliding down the slope to the minimum.
This same principle enables the optimization of parameters in systems that recommend products on e-commerce websites.
Now academic request are becoming passive-aggressive:
"Dear sir,
We tried contacting you several times, but since you never responded, we'd like to do so once more as a courtesy."
No, I am not interested in writing a 2-page opinion or mini-review.
Para el que diga que lo que se estudia en los primeros semestres no sirve para nada, acá estoy, 20 años después, escribiendo un código para transformar una lista de 1008 elementos en una matriz de 24x42 elementos. Es más complicado, pero esa es la idea. 🤓