DeepFusion is a web app that provides a graphical method for creating, training and testing neural networks. Unlike traditional graphical neural network modelers, DeepFusion has a node based editing system, which allows for much more customization, scaling and complexity. It also gives users the option to upload their own datasets for training and testing, allowing them to create a neural network that fit their needs. And as all the model and data generation is handled automatically, DeepFusion makes AI development incredibly easy without sacrificing its power or flexibility.
Team CommentsI chose to make this project because...
I chose to make this project as I found it very annoying to debug and develop neural networks traditionally. In code, network structures are defined by arbitrary numbers, making it impossible tell if the network you were making actually works or not until runtime, wasting precious time.What I found difficult and how I worked it out
I found it difficult to implement dynamic state updates in my project. Since DeepFusion needs to update every layer's IO dimension based on the current network structure in real time (for ease of development), I had to make several custom data structures and compute algorithms to make it possible.Next time, I would...
I would have connected this app to a compute backend server so that all the training can be offloaded to something more powerful. I would also try to design a better data management system next time, as it would be much easier to maintain and debug the program that way.
Very cool! Your presentation is funny, thoughtful, and clever. You made the concept of Deep Fusion fun to learn about and easier to understand, and your approach can be used to help others. Great work!