PinnedFlorian DebrauwerinTo cut a long paper shortSparrow | Improving alignment of dialogue agents via targeted human judgmentsProblem. Sparrow address the challenge of building safer conversational agent. More specifically, the problem consists of aligning the…Jan 3, 20231Jan 3, 20231
PinnedFlorian DebrauwerinTo cut a long paper shortAdam | A method for Stochastic OptimizationProblems. Training a neural network consist of optimizing stochastic functions in a high-dimensional space. In this context, finding a…Nov 29, 2022Nov 29, 2022
PinnedFlorian DebrauwerinTo cut a long paper shortBatchNorm | Accelerating Deep Network Training by Reducing Internal Covariate ShiftAddressing the covariate shift in deep networks greatly improves model training. Batch normalized network, when introduced, surpassed the…Nov 11, 2022Nov 11, 2022
PinnedFlorian DebrauwerinTo cut a long paper shortAlphaTensor | Discovering fast matrix multiplication algorithms with reinforcement learningImproving the efficiency of fundamental operations, such as matrix multiplications, significantly impacts AI systems. Automating the search…Oct 20, 2022Oct 20, 2022
PinnedFlorian DebrauwerinTo cut a long paper shortResNet | Deep Residual Learning for Image RecognitionProblems. Deep neural networks are susceptible to the degradation problem. While those deep networks start converging, further increasing…Aug 31, 2022Aug 31, 2022
Florian DebrauwerinTo cut a long paper shortEfficientNet | Rethinking Model Scaling for Convolutional Neural NetworksProblems. Convolutional Neural Networks are commonly scaled across depth, width, or resolution. Arbitrary scaling is tedious and often…Aug 12, 2022Aug 12, 2022
Florian DebrauwerinTo cut a long paper shortStyleGAN | A Style-Based Generator Architecture for Generative Adversarial NetworksProblem. SyleGAN is about understanding (and controlling) the image synthesis process in the generator of convolutional GANs. More…Jul 20, 2022Jul 20, 2022
Florian DebrauwerinTo cut a long paper shortDCGANs | Unsupervised Representation Learning with Deep Convolutional Generative Adversarial…Problem. This paper explores deep unsupervised CNN architecture, based on Generative Adversarial Networks (GAN). More specifically, it aims…Jul 13, 2022Jul 13, 2022
Florian DebrauwerinTo cut a long paper shortScaling Vision TransformersScale drives transformer’s performance. In this paper, Google Brain team explores the scaling properties of Vision transformer (ViT) across…Jun 26, 2022Jun 26, 2022
Florian DebrauwerinTo cut a long paper shortViT-Lite | Escaping the Big Data Paradigm with Compact TransformersTransformers have the reputation to be data-hungry NLP models. This paper shows that they can also be lightweights vision models and…Jun 6, 20221Jun 6, 20221
