Optimizing Neural Network Hyperparameters for Jet Feature Analysis
Explore the challenges faced during the training of neural networks for jet feature analysis, including issues with learning capabilities and flat losses. Solutions and progress updates are discussed through various hyperparameter adjustments and problem resolutions, ultimately aiming to improve model performance and address complex classification tasks.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
Updates 19.05.2023 Weekly meeting Greta Brianti
Fully connected test from 12.05.2023 updates Hyperparameter Trained for 30 epochs on 1k jets Numbers of layers 3 Hidden channels 256 Input features 410 Two problems: 1. The network doesn t learn over ~ 1.30 2. The losses are flat 2 20/03/2025 Greta Brianti - deeppp weekly meeting
Fully connected Hyperparameter Numbers of layers 4 Hidden channels 100 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k 3 20/03/2025 Greta Brianti - deeppp weekly meeting
Andrea dataset Hyperparameter Numbers of layers 4 Hidden channels 100 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k Input features 40 Classes 1 Loss BCE 4 20/03/2025 Greta Brianti - deeppp weekly meeting
Check on jet input variables 5 20/03/2025 Greta Brianti - deeppp weekly meeting
Only jet variables Hyperparameter Numbers of layers 4 Hidden channels 100 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k Input features 8 Classes 4 Loss CE 6 20/03/2025 Greta Brianti - deeppp weekly meeting
Only jet variables with binary class Hyperparameter Numbers of layers 4 Hidden channels 100 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k Input features 9 Classes 1 Loss BCE 7 20/03/2025 Greta Brianti - deeppp weekly meeting
Only jet variables with binary class First problem solved! Hyperparameter Numbers of layers 4 Hidden channels 100 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k Input features 9 Classes 1 Loss BCE 8 20/03/2025 Greta Brianti - deeppp weekly meeting
Losses flatness 9 jet features + 100 dummy features Hyperparameter Numbers of layers 4 Hidden channels 100 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k Input features 108 128 Classes 1 Loss BCE 9 jet features + 100 dummy features+ 20 low correlated features 9 20/03/2025 Greta Brianti - deeppp weekly meeting
Tracks management From -999 to nan array =np.full((40-len(curve_trk[i]),), np.nan,dtype=np.float64) Why? - Easier scaling When scaled: X[np.isnan(X)] = -999 10 20/03/2025 Greta Brianti - deeppp weekly meeting
All features 100 nodes Hyperparameter Numbers of layers 4 Hidden channels 100 250 Batch size 100 Learning rate 0.0001 Epochs 150 Dropout 0.5 Number of event 10k 250 nodes Input features 409 Classes 1 Loss BCE 11 20/03/2025 Greta Brianti - deeppp weekly meeting
QT Ntuples produced from the calibration group Run on grid with first event selection Download the ntuples from grid to ftag common folder Production of the file json with the dictionary of all the root file path Production from the root file the h5 file Data/MC plot production 12 20/03/2025 Greta Brianti - deeppp weekly meeting
QT Production from the root file the h5 file Data/MC plot production 13 20/03/2025 Greta Brianti - deeppp weekly meeting
Next steps and conclusion o Move to GNNs with binary label o For the QT: Debug 14 Greta Brianti - deeppp weekly meeting 20/03/2025
Next steps and conclusion o Move to GNNs with binary label o S o For the QT: Debug Thank you for your attention! 15 Greta Brianti - deeppp weekly meeting 20/03/2025
