MIT Video 1 (Paper 3 – Case study) for HL

Hello, this is Bndsbon’s blog post. In this blog post, I am going to explain the video which is part of the case study in the IB curriculum. I am also going to include a block diagram to illustrate my understanding of the video and connections with the case study.

The MIT video basically talks about the deep learning and convolutional neural networks. Incorporating them with the Case study “Autonomous Taxi”, I could explain the concepts and terminologies in Machine Learning to a better degree.

Neural Network

Also known as “stacked neural networks”, which is a network made of multiple layers, each layer is composed of different nodes. Frequent computation happens at the end of the nodes. 

Deep Learning

Also known as deep structured learning. It incorporates non-linear processing units for extraction and transformation. It can be unsupervised or supervised. It uses Gradient Descent and Back-Propagation.

For deep learning, the MIT professor talked about DeepTesla and DeepDriving as examples.

Here are several sensors used in the self-driving cars:

They gather the information which are thus sent to the CNNs ( convolutional neural networks)

These sensors help the vehicle to gather the information that is able to deduce its conditions in the following states:

Localization & Mapping: Where am I?

Scene Understanding: What is surrounding me?

Movement Planning: Where should I go?

Perceptron Algorithm

This algorithm is specially introduced in the MIT video

Here is its basic functions:

There are also a lot of other algorithms mentioned in the case study, such as the Dijkstra’s algorithm in the Taxi project, the nearest neighbor algorithm, etc.

Convolutional Neural Networks

They are Feedforward Artifical Neural Networks. Used mostly for analyzing visual images or photos. It is composed of three layers ( input layer, hidden layer, and output layers), the hidden layer usually has multiple layers including but not limited to pooling layers, convolutional layers.

In analyzing pictures, Pooling, CONV, RELU has been the major forces in helping the vehicles, autonomous vehicles, to identify the surrounding and make proper judgments in movements (momentum).

 

Here I have also included a Block diagram to guide our thinking in regarding with CNNs and sensors.

 

All in all, as AI and Google Autonomous cars are taking our sights day by day, it is important for us to realize what are the technologies behind, and incorporate them into our process of learning in IB CS.