The best articles of the week [21-27, Oct 2019]

2 minute read

The latest articles from the most prominent A.I scientists.

RESEARCH

“Fantastic Generalization Measures and Where to Find Them”: Interesting paper from Google, train over two thousand convolutional networks with systematic changes in commonly used hyperparameters. Hoping to uncover potentially causal relationships between each measure and generalization. Link
“Discrete Residual Flow for ProbabilisticPedestrian Behavior Prediction”: a convolutional neural network for human motion prediction that captures the uncertainty inherent in long-range motion forecasting. Link
“R-SQAIR: Relational Sequential Attend, Infer, Repeat”: sequential attention model can benefit from incorporating an explicit relational module which infers pairwise object interactions in parallel. Link
“Stabilizing transformers for reinforcement learning”: transformer … are too unstable to train in the RL setting. We presented a new architectural variant of the transformer model, the GTrXL, which has increased performance, more stable optimization, and greater robustness to initial seed and hyperparameters. Link
“Enhancing the transformer with explicit relational encoding for math problem solving”: incorporate Tensor-Product Representations within the Transformer in order to better support the explicit representation of relation structure … naturally learns to cluster symbol representations based on their structural position and relation to other symbols. Link

“Establishing an Evaluation Metric to quantify Climate Change Image Realism”: using a conditional generative model that illustrates the consequences of climate change-induced flooding to encourage public interest and awareness on the issue. Link
“Predicting ice flow using machine learning”: present IceNet dataset, use an unsupervised learning model to predict future ice flow. Link
“Dissecting racial bias in an algorithm used to manage the health of populations”: evidence of racial bias in one widely used algorithm, such that Black patients assigned the same level of risk by the algorithm are sicker than White patients. Link

Mathematics Dataset: generates mathematical question and answer pairs Link
“Icentia11K: An Unsupervised Representation Learning Dataset for Arrhythmia Subtype Discovery”: ECG dataset of continuous raw signals for representation learning containing 11 thousand patients and 2 billion labeled beats. Link