Industry use cases of Neural Networks
Let us understand what Kubernetes is with some follow-up questions and try answering them.
Asking question is the best way of learning and visualizing things. Lets start.
👉 What is a neuron? What is the most commonly used term of the neuron and why?
✔ NEURON -> Linear f(n) -> y(Yhat)=b+wx . commonly used term for neuron is PERCEPTRON => one neuron, one layer.
👉What is a loss in Neural Network and How NN calculate it?
✔|Yhat -Y| = Loss/cost — Absolute Square ~ 0 => model trained correctly.
👉 What is Initialiser? Name two types of Initialiser.
✔it initializes the random weight and bias for the NN model. 1. Kernel initializer (weight) 2. Bias initializer (bias)
👉 What is Optimiser?
✔ it helps to change weight and bias.
👉 What is Dense?
✔ used to create a layer of neurons .these layers are also called hidden layers.
👉What is Input, Output, and Hidden Layer?
✔ input — the first layer of NN
Hidden Layer — created with the help of Dense function to train the model
output — the last layer of NN
👉 What is the pre-trained weight?rut
✔after training, we can save a model as a .h5 file and can use the pre-trained weights for future predictions.
👉Which function did we use in NN to work?
✔ Sequential() function is used mostly
let’s visualize the NN.
NN consists of three main layers:
Input layer: The first layer of NN
Hidden layer: the layers used to train the model
Output layer: the last layer of NN.
.NEURON is a linear function that helps calculates the loss(the difference between the actual and predicted data) and if the loss is close to zero, the model is trained accurately.
If not, Backpropagation takes place and the Optimizer re-changes the weight and bias and feedforward. Loss function measures the loss and the process keeps going on until the loss is minimized.
Now, we had a basic intro to Neural Networks, let’s continue forward with some real use cases
Self-Driving Cars
It is Using Billions of Miles to Train Neural Networks
5 million miles a day extrapolates to 5.4 billion miles a year, or 200x more than Waymo’s expected to total a year from now. Tesla’s fleet is also growing by approximately 5,000 cars per week.
There are three key areas where data makes a difference:
- Computer vision
Whenever encounter an obstacle or object, the cameras will take a snapshot, which will be uploaded later over wifi. - Prediction
Tesla’s Director of AI, Andrej Karpathy, says the reason why nobody has achieved full autonomy “is that today’s software is not good enough to predict the future.” Whereas images used to train object detection require human labeling, a prediction neural network can learn correlations between past and future just from temporal sequences of events.
- Path planning/driving policy
The training process is similar to how a neural network learns to predict the behavior of other road users by drawing correlations between past and future.
How tesla leaps this problem?
Based on Karpathy’s comments about predicting cut-ins, Tesla can trigger a car to save a replay when it fails to correctly predict whether a vehicle ahead will cut into the Tesla’s lane. Similarly, Tesla may capture replay data when a neural network involved in path planning or driving policy fails to correctly predict the Tesla driver’s actions. Elon Musk has alluded to this capability (or something similar) in the past, although it’s not clear if it’s currently running in Tesla cars.
I hope you find this article interesting and do search more to get more astonishing facts and processes about NN.
Thank you for reading.
Do connect me on Linkedin.