Please enable JavaScript.

Coggle requires JavaScript to display documents.

Case Study (Terminology (Greedy Algorithm (Example: Nearest neighbor…

- - - - Difference between Dikstra and the Travelling salesman problem
        
        The nearest neighbor algorithm is only concerned with the most optimal path between two nodes that are relatively close to each other. (not the whole path Applying this across many nodes will give us a good approximation of a short route, but there is probably a more efficient solution
        
        Dijkstra's algorithm is concerned about finding the optimal path between two vertices in a graph. Only finding the shortest distance between nodes and comparing adjacent nodes, but doesn't require a path that encompasses every vertex. Travelling salesman problem entails visiting every single node to find the shortest route
- - - - Sensor modules are cheaper because no processing is done.
      - Electronic Control unit has all the data required from the sensors. More sensors can be deployed
    - - ECU can be very expensive as it needs high processing power
      - Sensor modules require a wide band with to transfer large amounts of data to a central location
  - - - Lower bandwidth: resulting in a cheaper interface
      - The ECU does not need to have high processing power because it is only object data. Adding a sensor does not require a significant increase in power because most of the processing can be handled by the new sesnor
    - - Sensor's require an application processor resulting in higher costs per sensor and also larger sensors as well
      - The ECU (Electrical Control unit) / central processor does not have access to all of the data. Only object data, not actual sensor data
- - - - Why TF Is this important
        
        Vehicles are essentially able to talk to each other
      - Advantages :check:
        
        Reduced accidents caused by human error
        
        Provide intersection assistance: awareness to blind spots
        
        More efficient gas mileage
      - Disadvantages :red_cross:
        
        Would take a long time to implement
        
        Costly to install
  - - - Why is it important
        
        Provide cars with real time information on road conditions and vice versa for better management of traffic
      - Advantages :check:
        
        Less congestion due to more efficient traffic management
        
        More efficient mileage due to less time spent in traffic
      - Disadvantages :red_cross:
        
        Costly to set up
        
        Requires large wide spread implementation for the system to be efficient
- - - - Filter or kernel (in image processing) is a small matrix which is used for blurring, sharpening, edge detection and more
        
        These learnable filters convolve the image by computing the dot product of the input dimensions it is given. E.g 3.x 3 filter takes the dot product of a 3 x 3 pixels. This will produce a 2-dimensional activation map, which get stacked.
        
        Filter stride: The number of pixels the filter moves per iteration
        
        Effect: it decides the output volume for the next layer
      - As we venture deeper in the neural network. The filters gradually become more sophisticated. E.g Detecting circles to birds
      - Receives some sort of input, processes the input in some way, and outputs the processed output to the next layer (adjacent layer)
      - Each convolutional layer has a specified number of layers (kernels) to recognise patterns. I.e one filter could detect circles in an image.
      - Four main layers Convolutional Layer
        
        Pooling Layer
        
        Gradually reduces the spatial size of the input and reduce the amount of parameters and computation in the network. It usually uses MAX POOLING. It allows assupmtoins to be made and can help with overfitting
        
        It will take a kernel and slide it over an image with a stride that ensures overlapping is prevented. For each region represented by the filter the maximum of that region will be taken and used in a new output matrix. I.e these max statistics are then used to describe the image
        
        Convolutional Layer
        
        A layer that essentially does the heavy lifting of the program. These layers contain learner filters/kernels #
        
        Activation layer / Rectified linear units layer (RELU layer)
        
        A layer that introduces non-linearity to the convolutional layer because most relationships in the real world are non-linear. This is done by changing all the negative activations to zero. This also somewhat rectifies the gradient problem.
        
        Fully Connected Layer
        
        Neurons in a fully connected layer that have full connections to every activation in the previous layer
- - - - An artificial neural network that has 3 or more layers (1 input layer, 1 output layer, and 1+ hidden layer) Each Layer contains multiple nodes. Adjacent layers are joined by connections
        
        Input Node: Does no computation, provide information from the outside world to the network
        
        Hidden node: A layer that has no interaction with the outside world. Performs computation on data received from the input nodes and transfers it to the output nodes
        
        Output Nodes: Responsible for computation to transfer information from the network to the outside world.