A variant of LinUCB bandit algorithm with local differential privacy guarantee

Last update: Oct 25, 2022

Related tags

Overview

LDP LinUCB Description
Model Architecture
Dataset
Environment Requirements
Script Description
Model Description
- Performance
Description of Random Situation
ModelZoo Homepage

LDP LinUCB

Locally Differentially Private (LDP) LinUCB is a variant of LinUCB bandit algorithm with local differential privacy guarantee, which can preserve users' personal data with theoretical guarantee.

Paper: Kai Zheng, Tianle Cai, Weiran Huang, Zhenguo Li, Liwei Wang. "Locally Differentially Private (Contextual) Bandits Learning." Advances in Neural Information Processing Systems. 2020.

Model Architecture

The server interacts with users in rounds. For a coming user, the server first transfers the current model parameters to the user. In the user side, the model chooses an action based on the user feature to play (e.g., choose a movie to recommend), and observes a reward (or loss) value from the user (e.g., rating of the movie). Then we perturb the data to be transferred by adding Gaussian noise. Finally, the server receives the perturbed data and updates the model. Details can be found in the original paper.

Dataset

Note that you can run the scripts based on the dataset mentioned in original paper. In the following sections, we will introduce how to run the scripts using the related dataset below.

Dataset used: MovieLens 100K

Dataset size：5MB, 100,000 ratings (1-5) from 943 users on 1682 movies.
Data format：csv/txt files

Environment Requirements

Hardware (Ascend/GPU)
- Prepare hardware environment with Ascend or GPU processor.
Framework
- MindSpore
For more information, please check the resources below：
- MindSpore Tutorials
- MindSpore Python API

Script Description

Script and Sample Code

├── model_zoo
    ├── README.md                                // descriptions about all the models
    ├── research
        ├── rl
            ├── ldp_linucb
                ├── README.md                    // descriptions about LDP LinUCB
                ├── scripts
                │   ├── run_train_eval.sh        // shell script for running on Ascend
                ├── src
                │   ├── dataset.py               // dataset for movielens
                │   ├── linucb.py                // model
                ├── train_eval.py                // training script
                ├── result1.png                  // experimental result
                ├── result2.png                  // experimental result

Script Parameters

Parameters for preparing MovieLens 100K dataset

'num_actions': 20         # number of candidate movies to be recommended
'rank_k': 20              # rank of rating matrix completion

Parameters for LDP LinUCB, MovieLens 100K dataset

'epsilon': 8e5            # privacy parameter
'delta': 0.1              # privacy parameter
'alpha': 0.1              # failure probability
'iter_num': 1e6           # number of iterations

Launch

running on Ascend

python train_eval.py > result.log 2>&1 &

The python command above will run in the background, you can view the results through the file result.log.

The regret value will be achieved as follows:

--> Step: 0, diff: 348.662, current_regret: 0.000, cumulative regret: 0.000
--> Step: 1, diff: 338.457, current_regret: 0.000, cumulative regret: 0.000
--> Step: 2, diff: 336.465, current_regret: 2.000, cumulative regret: 2.000
--> Step: 3, diff: 327.337, current_regret: 0.000, cumulative regret: 2.000
--> Step: 4, diff: 325.039, current_regret: 2.000, cumulative regret: 4.000
...

Model Description

The original paper assumes that the norm of user features is bounded by 1 and the norm of rating scores is bounded by 2. For the MovieLens dataset, we normalize rating scores to [-1,1]. Thus, we set sigma in Algorithm 5 to be $$4/epsilon * sqrt(2 * ln(1.25/delta))$$.

Performance

The performance for different privacy parameters:

x: number of iterations
y: cumulative regret

The performance compared with optimal non-private regret O(sqrt(T)):

x: number of iterations
y: cumulative regret divided by sqrt(T)

Description of Random Situation

In train_eval.py, we randomly sample a user at each round. We also add Gaussian noise to the date being transferred.

ModelZoo Homepage

Please check the official homepage.

You might also like...

Open source home automation that puts local control and privacy first

Home Assistant Open source home automation that puts local control and privacy first. Powered by a worldwide community of tinkerers and DIY enthusiast

57k Jan 2, 2023

Open source home automation that puts local control and privacy first.

Home Assistant Open source home automation that puts local control and privacy first. Powered by a worldwide community of tinkerers and DIY enthusiast

57k Jan 1, 2023

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis. You write a high level configuration file specifying your in

917 Jan 3, 2023

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis. You write a high level configuration file specifying your in

915 Dec 29, 2022

Author's PyTorch implementation of TD3+BC, a simple variant of TD3 for offline RL

A Minimalist Approach to Offline Reinforcement Learning TD3+BC is a simple approach to offline RL where only two changes are made to TD3: (1) a weight

193 Dec 23, 2022

Official PyTorch code for Mutual Affine Network for Spatially Variant Kernel Estimation in Blind Image Super-Resolution (MANet, ICCV2021)

Mutual Affine Network for Spatially Variant Kernel Estimation in Blind Image Super-Resolution (MANet, ICCV2021) This repository is the official PyTorc

139 Dec 29, 2022

A playable version of Chess – classic two-player, various AI levels, and the crazyhouse variant! Written in Python 3

A playable version of Chess – classic two-player, various AI levels, and the crazyhouse variant! Written in Python 3. Requires the installation of PIL/Pillow and Requests

1 Dec 24, 2021

Minimalistic generic chess variant GUI using pyffish and PySimpleGUI, based on the PySimpleGUI Chess Demo

FairyFishGUI Minimalistic generic chess variant GUI using pyffish and PySimpleGUI, based on the PySimpleGUI Chess Demo. Supports all chess variants su

6 Dec 20, 2022

A variant caller for the GBA gene using WGS data

Gauchian: WGS-based GBA variant caller Gauchian is a targeted variant caller for the GBA gene based on a whole-genome sequencing (WGS) BAM file. Gauch

16 Oct 13, 2022

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis. You write a high level configuration file specifying your in

917 Jan 3, 2023

Implementation of the Transformer variant proposed in "Transformer Quality in Linear Time"

FLASH - Pytorch Implementation of the Transformer variant proposed in the paper Transformer Quality in Linear Time Install $ pip install FLASH-pytorch

209 Dec 28, 2022

Pipenv-local-deps-repro - Reproduction of a local transitive dependency on pipenv

Reproduction of the pipenv bug with transitive local dependencies. Clone this re

2 Jan 11, 2022

A simple python script to dump remote files through a local file read or local file inclusion web vulnerability.

A simple python script to dump remote files through a local file read or local file inclusion web vulnerability. Features Dump a single file w

48 Dec 3, 2022

Official code for Score-Based Generative Modeling through Stochastic Differential Equations

Score-Based Generative Modeling through Stochastic Differential Equations This repo contains the official implementation for the paper Score-Based Gen

818 Jan 6, 2023

Code for "Infinitely Deep Bayesian Neural Networks with Stochastic Differential Equations"

Infinitely Deep Bayesian Neural Networks with SDEs This library contains JAX and Pytorch implementations of neural ODEs and Bayesian layers for stocha

95 Nov 26, 2021

Supplementary code for the paper "Meta-Solver for Neural Ordinary Differential Equations" https://arxiv.org/abs/2103.08561

Meta-Solver for Neural Ordinary Differential Equations Towards robust neural ODEs using parametrized solvers. Main idea Each Runge-Kutta (RK) solver w

25 Aug 12, 2021

PyTorch implementation for SDEdit: Image Synthesis and Editing with Stochastic Differential Equations

SDEdit: Image Synthesis and Editing with Stochastic Differential Equations Project | Paper | Colab PyTorch implementation of SDEdit: Image Synthesis a

536 Jan 5, 2023

Partial implementation of ODE-GAN technique from the paper Training Generative Adversarial Networks by Solving Ordinary Differential Equations

ODE GAN (Prototype) in PyTorch Partial implementation of ODE-GAN technique from the paper Training Generative Adversarial Networks by Solving Ordinary

15 Feb 10, 2022

Leibniz is a python package which provide facilities to express learnable partial differential equations with PyTorch

Beijing ColorfulClouds Technology Co.,Ltd.

16 Aug 7, 2022

A variant of LinUCB bandit algorithm with local differential privacy guarantee

Related tags

Overview

Contents

You might also like...

Open source home automation that puts local control and privacy first

Open source home automation that puts local control and privacy first.

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Author's PyTorch implementation of TD3+BC, a simple variant of TD3 for offline RL

Official PyTorch code for Mutual Affine Network for Spatially Variant Kernel Estimation in Blind Image Super-Resolution (MANet, ICCV2021)

A playable version of Chess – classic two-player, various AI levels, and the crazyhouse variant! Written in Python 3

Minimalistic generic chess variant GUI using pyffish and PySimpleGUI, based on the PySimpleGUI Chess Demo

A variant caller for the GBA gene using WGS data

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Implementation of the Transformer variant proposed in "Transformer Quality in Linear Time"

Pipenv-local-deps-repro - Reproduction of a local transitive dependency on pipenv

A simple python script to dump remote files through a local file read or local file inclusion web vulnerability.

Official code for Score-Based Generative Modeling through Stochastic Differential Equations

Code for "Infinitely Deep Bayesian Neural Networks with Stochastic Differential Equations"

Supplementary code for the paper "Meta-Solver for Neural Ordinary Differential Equations" https://arxiv.org/abs/2103.08561

PyTorch implementation for SDEdit: Image Synthesis and Editing with Stochastic Differential Equations

Partial implementation of ODE-GAN technique from the paper Training Generative Adversarial Networks by Solving Ordinary Differential Equations

Leibniz is a python package which provide facilities to express learnable partial differential equations with PyTorch

Releases(v1.1.0)

v1.1.0(Jun 6, 2022)

v1.0.0(Jun 6, 2022)

Owner

Weiran Huang

Data cleaning tools for Business analysis

Code for the DH project "Dhimmis & Muslims – Analysing Multireligious Spaces in the Medieval Muslim World"

Full ELT process on GCP environment.

Gaussian processes in TensorFlow

Python dataset creator to construct datasets composed of OpenFace extracted features and Shimmer3 GSR+ Sensor datas

Python utility to extract differences between two pandas dataframes.

MIR Cheatsheet - Survival Guidebook for MIR Researchers in the Lab

This is an example of how to automate Ridit Analysis for a dataset with large amount of questions and many item attributes

Python Project on Pro Data Analysis Track

Snakemake workflow for converting FASTQ files to self-contained CRAM files with maximum lossless compression.

EOD Historical Data Python Library (Unofficial)

Sensitivity Analysis Library in Python (Numpy). Contains Sobol, Morris, Fractional Factorial and FAST methods.

Synthetic data need to preserve the statistical properties of real data in terms of their individual behavior and (inter-)dependences

SNV calling pipeline developed explicitly to process individual or trio vcf files obtained from Illumina based pipeline (grch37/grch38).

Additional tools for particle accelerator data analysis and machine information

Pipetools enables function composition similar to using Unix pipes.

PyClustering is a Python, C++ data mining library.

Pipeline to convert a haploid assembly into diploid

A Pythonic introduction to methods for scaling your data science and machine learning work to larger datasets and larger models, using the tools and APIs you know and love from the PyData stack (such as numpy, pandas, and scikit-learn).

This tool parses log data and allows to define analysis pipelines for anomaly detection.