100 likes | 170 Views
Dive into Lucas Divine's neural network project for music albums evaluation. Explore the implementation of a feed-forward, acyclic, multilayer perceptron aiming to determine the worthiness of purchasing a CD based on various factors like price, number of discs, song statistics, and user's gut feeling. Discover the importance of static weights in ensuring fair judgements across different CDs. Check out sample results that showcase the network's ability to provide recommendations on CDs. For a detailed insight, refer to the project report (musicNN.pdf) and test the network using musicNeuralNetwork.m in Matlab.
E N D
Should I Buy That CD? A Neural Network Project By Lucas Divine
The Problem • If I pick a certain Music Album, I would like a neural net to tell me if it is worth buying or listening to. Motivation • To create a neural network that is comprehensible. A neural network that functions as closely as possible to how biological neural networks would.
The Neural Network • Designed as a feed-forward, acyclic, multilayer perceptron with no back propagation of errors. • Hyperbolic tangent as the activation function to prevent any specific data from skewing the overall results.
1 Price [1] Facts 2 [2] # Discs Opinion Buy Or Pass 3 # Songs [3] Stats User’s Gut Feeling 12 [12] Data Preprocessing Hidden Layer Output Layer Input Layer Diagram of the Network
Gut Feeling Input • An input that affects how a majority of people decide on CDs • Affect the network results as it should • Allows for user interaction in the network
Weights • Most crucial step in the process • Static weights to affect all CDs the same way, thus judging fairly across the board • Made sure that all data had a scaled effect on the outcome depending on relevance
These Results • Vary quite a bit (so does the data) • Are specifically affected by varying stats • Show that the ANN does tell you to BUY good CDs and SORRY for bad ones • A ‘linear’ pattern does not occur, thus demonstrating the non-linear element of the perceptron network • User ‘gut feeling’ input has the effect that it should
Thank you! • Please refer to the project report (musicNN.pdf) for a deeper understanding • Also run musicNeuralNetwork.m in matlab to test it out yourself