Papers I Read Notes and Summaries

Representation Tradeoffs for Hyperbolic Embeddings

Introduction

  • The paper describes a combinatorial approach to embed trees into hyperbolic spaces without performing optimization.

  • ... Continue reading

Learned Optimizers that Scale and Generalize

Introduction

  • The paper introduces a learned gradient descent optimizer that has low memory and computational overhead and... Continue reading


One-shot Learning with Memory-Augmented Neural Networks

Introduction

  • The paper demonstrates that Memory Augmented Neural Networks (MANN) are suitable for one-shot learning by introducing... Continue reading


BabyAI - First Steps Towards Grounded Language Learning With a Human In the Loop

Introduction

  • BabyAI is a research platform to investigate and support the feasibility of including humans in the... Continue reading


Poincaré Embeddings for Learning Hierarchical Representations

Introduction

  • Much of the work in representation leaning uses Euclidean vector spaces to embed datapoints (like words,... Continue reading


When Recurrent Models Don’t Need To Be Recurrent

Introduction

  • The paper explores “if a well behaved RNN can be replaced by a feed-forward network of... Continue reading


HoME - a Household Multimodal Environment

Introduction

  • Environment for learning using modalities like vision, audio, semantics, physics and interaction with objects and other... Continue reading


Emergence of Grounded Compositional Language in Multi-Agent Populations

Introduction

  • The paper provides a multi-agent learning environment and proposes a learning approach that facilitates the emergence... Continue reading


A Semantic Loss Function for Deep Learning with Symbolic Knowledge

Introduction

  • The paper proposes an approach for using symbolic knowledge in deep learning systems. These constraints are... Continue reading


Hierarchical Graph Representation Learning with Differentiable Pooling

Introduction

  • Most existing GNN (Graph Neural Network) methods are inherently flat and are unable to process the... Continue reading