Papers I Read Notes and Summaries

To Tune or Not to Tune? Adapting Pretrained Representations to Diverse Tasks

Model Primitive Hierarchical Lifelong Reinforcement Learning


  • The paper presents a framework that uses diverse suboptimal world models that can be used to... Continue reading

TuckER - Tensor Factorization for Knowledge Graph Completion


  • TuckER is a simple, yet powerful linear model that uses Tucker decomposition for the task of... Continue reading

Linguistic Knowledge as Memory for Recurrent Neural Networks

Diversity is All You Need - Learning Skills without a Reward Function


  • The paper proposes an approach to learn useful skills without a reward function by maximizing an... Continue reading

Modular meta-learning


  • The paper proposes an approach for learning neural networks (modules) that can be combined in different... Continue reading

Hierarchical RL Using an Ensemble of Proprioceptive Periodic Policies


  • The paper proposes a simple and robust approach for hierarchically training an agent in the sparse... Continue reading

Efficient Lifelong Learning with A-GEM


  • A new (and more realistic) evaluation protocol for lifelong learning where each data point is observed... Continue reading

Pre-training Graph Neural Networks with Kernels


  • The paper proposes a pretraining technique that can be used with the GNN architecture for... Continue reading

Smooth Loss Functions for Deep Top-k Classification


  • For top-k classification tasks, cross entropy is widely used as the learning objective even though it... Continue reading