AI Insights-In a Minute - Pilot

1 minute read

Published:

Short Summaries of AI Research.

In 2023, I plan to increase my capacity for consuming research papers. For better retention of all the research, I realized that it is important to have an organized way of summarizing the papers and to compile these summaries in a central location. After writing a few such paper summaries, I thought it might be a good idea to share them publicly, hoping that in the long run they might be useful to me and others.

With this in mind, I am starting a YouTube channel called “AI Insights — In a Minute” and a GitHub repository called “ML Papers Intuitions,” which will feature brief summaries of the latest machine learning research papers. These summaries will include a brief background of the research, the key idea, and one notable result. While they will not be sufficient for understanding the papers in detail, they should provide enough information to determine whether a full read is worthwhile and give readers a sense of the paper’s direction.

I have already shared four videos on Deep Learning approaches for Time Series Forecasting, an area I have been working on recently. I will continue to maintain the channel if found useful by others. I welcome any feedback as I continue to create more summaries.

Links to pilot videos:

  1. A quick review of Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecast
  2. A quick review of N-BEATS: Neural basis expansion analysis for interpretable time series forecasting
  3. A quick review of N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting
  4. A quick review of Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence

Link to GitHub repo with summaries of the same papers:

ML Papers Intuitions

You May Also Enjoy

Transfer Learning Using MXNet

30 minute read

Published:

This post provides you an easy to follow tutorial on how to “train a base neural net” on a dataset and use that pre-trained network to “transfer learn” on a different dataset using MXNet/Gluon framework.

Getting Deeper into Categorical Encodings for Machine Learning

less than 1 minute read

Published:

Jumping from simple algorithms to complex ones does not always boost performance if the feature engineering is not done right. One type of features that do not easily give away the information they contain are categorical features.

Custom Loss Functions for Gradient Boosting

less than 1 minute read

Published:

Optimize what matters — Authors: Prince Grover and Sourav Dey. This post is our attempt to summarize the importance of custom loss functions in many real-world problems — and how to implement them with the LightGBM gradient boosting package.