Infinite Recommendation Networks (∞-AE)

This repository contains the implementation of ∞-AE from the paper “Infinite Recommendation Networks: A Data-Centric Approach” [arXiv] where we leverage the NTK of an infinitely-wide autoencoder for implicit-feedback recommendation. Notably, ∞-AE:

• Is easy to implement (<50 lines of relevant code)
• Has a closed-form solution
• Has only a single hyper-parameter, $\lambda$
• Even though simplistic, outperforms all complicated SoTA models

The paper also proposes Distill-CF: how to use ∞-AE for data distillation to create terse, high-fidelity, and synthetic data summaries for model training. We provide Distill-CF’s code in a separate GitHub repository.

If you find any module of this repository helpful for your own research, please consider citing the below under-review paper. Thanks!

@article{sachdeva2022b,
  title={Infinite Recommendation Networks: A Data-Centric Approach},
  author={Sachdeva, Noveen and Dhaliwal, Mehak Preet and Wu, Carole-Jean and McAuley, Julian},
  journal={arXiv preprint arXiv:2206.02626},
  year={2022}
}

Code Author: Noveen Sachdeva ([email protected])

Setup

Environment Setup

pip install -r requirements.txt

Data Setup

Once you’ve correctly setup the python environment, the following script will download the ML-1M dataset and preprocess it for usage:

./prep.sh

How to train ∞-AE?

• Edit the hyper_params.py file which lists all config parameters of ∞-AE.
• Finally, type the following command to train and evaluate ∞-AE:

CUDA_VISIBLE_DEVICES=0 python main.py

Results sneak-peak

Below are the nDCG@10 results for the datasets used in the paper:

Note: ∞-AE’s results on the Netflix dataset (marked with a *) are obtained by training only on 5% of the total users. Note however, all other methods are trained on the full dataset.

MIT License

GitHub

View Github