Today is the day that the new T53 input-vector weight regulator fires! That's a lot of words, so to explain what happens today and why it's so significant, I'll break it down piece by piece.
T53:
The T53 engine a powerful turboshaft engine developed by Lycoming that powers everything from the Huey to the Cobra. Because I am a nerd, every program that I write in this project phase is named after a Lycoming engine where each new program takes the name of a more powerful engine. Naming programs after engines started with the previous phase of the project, the Lister phase. This name was chosen by suggestion of my father who told me about how the Lister company made small one cylinder engines that while simple, performed well. This was a fitting name for programs that only functioned to predict simple sine curves with layered noise and had the goal of tracking the overall trend through the noise.
Input-Vector:
This one is pretty easy. The input-vector is just a fancy way of saying what the network sees. If you want to be even more fancy (unbearable), you could call it an input tensor.
Weight Regulator:
This may sound like a cool new fitness tracker app but in reality weight regulation in this context is the term for how you weight the importance of the various parts of the input-vector. Not all inputs are created equal, some can be very helpful while others will downright hurt your prediction accuracy. By changing the weights on inputs our networks can tune in to the more helpful data inputs and find what is best.
So what does this all mean? It means that today is the day in which the cool new advanced T53 will crank through possible input-vector weights and learn and adapt as it goes until it finds the ideal inputs. So far it has been able to boost performance by 10% (percentage points) in directional accuracy prediction. This is an exciting development and I look forward to using it as the project moves forwards.
