Papers To Read
- BREAKING THE SOFTMAX BOTTLENECK: A HIGH-RANK RNN LANGUAGE MODEL
- Visual Curiosity: Learning to Ask Questions to Learn Visual Recognition
- TARMAC: TARGETED MULTI-AGENT COMMUNICATION
- Towards Understanding Linear Word Analogies
- Understanding the impact of entropy on policy optimization
- Do explanations make VQA models more predictable to a human?
- How agents see things: On visual representations in an emergent language game
- Semantic Parsing for Task Oriented Dialog using Hierarchical Representations
- [https://openreview.net/pdf?id=ryQu7f-RZ] AmsGrad
Projects Roadmap
-
Visdial Research
RNN in Pytorch HRED in Pytorch LMFUSION in Pytorch Topical Hred Augmenting Neural Response Generation with Context-Aware Topical Attention A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialogues Implemention of Co-Operating Games GuessWhat?! Visual object discovery through multi-modal dialogue
-
Presentation for PRML group
Reading the VI paper Reading the VAE Paper 0. [https://arxiv.org/abs/1312.6114] Original Paper 1. [https://arxiv.org/abs/1606.05908] Tutorial 2. Implementation in Pytorch(Ipython notebook prefered for tutorial) VARIATIONAL INFERENCE: FOUNDATIONS AND INNOVATIONS Tutorial
-
Fashion Image Tagging
Reading the Google open images [] Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labels (Tags Riken Noisy Fashion) Masking: A New Perspective of Noisy Supervision (Tags Riken Noisy Fashion)
-
Autograd & Computational Graph
Mostly Cuda Learning
- Efficient Large-scale Approximate Nearest Neighbor Search on OpenCL FPGA
- Billion-scale similarity search with GPUs FAIR (Tags CUDA FAISS LOPQ Image Search)
- Locally Optimized Product Quantization for Approximate Nearest Neighbor Search (Tags CUDA LOPQ Image Search)
- Sparse Tensor tutorial (Tags SparseTensor CUDA OPENAI )
RL Resources
*Papers to Implement
- Poincaré Embeddings for Learning Hierarchical Representations
- A DISCIPLINED APPROACH TO NEURAL NETWORK HYPER-PARAMETERS: PART 1 – LEARNING RATE, BATCH SIZE, MOMENTUM, AND WEIGHT DECAY
- https://github.com/rusty1s/pytorch_geometric.git
Blogs to Read
- Hovrod
- [https://towardsdatascience.com/how-to-build-a-gated-convolutional-neural-network-gcnn-for-natural-language-processing-nlp-5ba3ee730bfb] Gated Conv Neural Network
- [https://www.fast.ai/2018/07/02/adam-weight-decay/] Super Convergence
- [http://www.phontron.com/class/nn4nlp2017/schedule.html] NLP course
- [https://stats.stackexchange.com/questions/281240/why-is-the-cost-function-of-neural-networks-non-convex] Why loss function is convex still the loss i s surface is not convex