INTELlinext: A Fully Integrated LSTM and HMM-Based Solution for Next-App Prediction With Intel SUR SDK Data Collection

Cyril Gorlla; Jared Thach; Hiroki Hoshida

Communication Dans Un Congrès Année : 2022

INTELlinext: A Fully Integrated LSTM and HMM-Based Solution for Next-App Prediction With Intel SUR SDK Data Collection

(1) , (1) , (1)

Cyril Gorlla

Fonction : Auteur
PersonId : 1097911

Halıcıoğlu Data Science Institute

Jared Thach

Fonction : Auteur
PersonId : 1123281

Halıcıoğlu Data Science Institute

Hiroki Hoshida

Fonction : Auteur
PersonId : 1123282

Halıcıoğlu Data Science Institute

Résumé

As the power of modern computing devices increases, so too do user expectations for them. Despite advancements in technology, computer users are often faced with the dreaded spinning icon waiting for an application to load. Building upon our previous work developing data collectors with the Intel System Usage Reporting (SUR) SDK, we introduce INTELlinext, a comprehensive solution for next-app prediction for application preload to improve perceived system fluidity. We develop a Hidden Markov Model (HMM) for prediction of the k most likely next apps, achieving an accuracy of 70% when k = 3. We then implement a long short-term memory (LSTM) model to predict the total duration that applications will be used. After hyperparameter optimization leading to an optimal lookback value of 5 previous applications, we are able to predict the usage time of a given application with a mean absolute error of ~45 seconds. Our work constitutes a promising comprehensive application preload solution with data collection based on the Intel SUR SDK and prediction with machine learning.

Mots clés

Neural networks Markov decision processes Active learning settings lstm hidden markov model app prediction

Domaines

Apprentissage [cs.LG]

Fichier principal

sample-sigplan.pdf (1.04 Mo)

download.png (36.29 Ko)

All_Inputs.png (104.85 Ko)

Cursor_States.png (48.53 Ko)

Foreground.png (49.64 Ko)

Foreground_Scatter.png (57.88 Ko)

Mouse_X.png (26.74 Ko)

Mouse_X_Scatter.png (199.35 Ko)

Mouse_Y_Scatter.png (224.54 Ko)

User_Scatter.png (49.78 Ko)

User_Wait.png (26.03 Ko)

desktop_fg.png (74.11 Ko)

laptop_fg.png (89.56 Ko)

logo.png (81.22 Ko)

lstm.png (87.08 Ko)

lstm_diagram.png (48.85 Ko)

lstm_loss.png (69.46 Ko)

notepad.png (65.81 Ko)

t10.png (287.5 Ko)

Origine	Fichiers produits par l'(les) auteur(s)

Format	Figure, Image
Origine	Fichiers produits par l'(les) auteur(s)

Cyril Gorlla : Connectez-vous pour contacter le contributeur

https://hal.science/hal-03825175

Soumis le : samedi 22 octobre 2022-01:28:04

Dernière modification le : mardi 19 novembre 2024-14:04:09

Dates et versions

hal-03825175 , version 1 (22-10-2022)

Identifiants

HAL Id : hal-03825175 , version 1

Citer

Cyril Gorlla, Jared Thach, Hiroki Hoshida. INTELlinext: A Fully Integrated LSTM and HMM-Based Solution for Next-App Prediction With Intel SUR SDK Data Collection. Halıcıoğlu Data Science Institute Capstone Showcase, Apr 2022, La Jolla, United States. ⟨hal-03825175⟩

Exporter

BibTeX XML-TEI Dublin Core DC Terms EndNote DataCite

49 Consultations

87 Téléchargements

INTELlinext: A Fully Integrated LSTM and HMM-Based Solution for Next-App Prediction With Intel SUR SDK Data Collection

Résumé

Mots clés

Domaines

Dates et versions

Identifiants

Citer

Exporter

Partager