DeepFix
Jump to navigation
Jump to search
See: DeepFix Program Repair System, DeepFix Seq2Seq Neural Network, DeepFix Convolutional Neural Network.
References
2017a
- (Gupta et al., 2017) ⇒ Rahul Gupta, Soham Pal, Aditya Kanade, and Shirish Shevade. (2017). “DeepFix: Fixing Common C Language Errors by Deep Learning.” In: Proceeding of AAAI.
- QUOTE: We present an end-to-end solution, called DeepFix, that does not use any external tool to localize or fix errors. We use a compiler only to validate the fixes suggested by DeepFix. At the heart of DeepFix is a multi-layered sequence-to-sequence neural network with attention (Bahdanau, Cho, and Bengio 2014), comprising of an encoder recurrent neural network (RNN) to process the input and a decoder RNN with attention that generates the output. The network is trained to predict an erroneous program location along with the correct statement. DeepFix invokes it iteratively to fix multiple errors in the program one-by-one.
2017b
- (Kruthiventi et al., 2017) ⇒ Srinivas S. S. Kruthiventi, Kumar Ayush, and R. Venkatesh Babu. (2017). “DeepFix: A Fully Convolutional Neural Network for Predicting Human Eye Fixations.” In: IEEE Transactions on Image Processing Journal, 26(9). doi:10.1109/TIP.2017.2710620
- QUOTE: In this work, we propose a fully convolutional neural network - DeepFix, for predicting human eye fixations on images in the form of a saliency map. Our model, inspired from VGG net [20], is a very deep network with 20 convolutional layers, each of a small kernel size, operating in succession on an image. The network is designed to capture object-level semantics, which can occur at multiple scales, efficiently through inception style [21] convolution blocks. Each inception module consists of a set of convolution layers with different kernel sizes operating in parallel. The global context of the scene, which is crucial for saliency prediction, is captured using convolutional layers with very large receptive fields. These layers are placed towards the end of the network and replace the densely connected inner product layers commonly present in convolutional net.