Resource Optimization for Facial Recognition Systems
Facial recognition systems have found substantial applications in various industries, including security, marketing, and healthcare. Nevertheless, their real-time applications pose significant computational challenges. This thesis explores the strengths and limitations of different machine learning algorithms: UCB-1, SW-UCB, D-UCB, and LSTM in optimising resource allocation and traffic flow prediction for facial recognition systems. Through an empirical analysis, it was found that UCB-1, while computationally efficient and suitable for quick decision-making, falls short in non-stationary environments due to its assumption of stationary reward distributions and its inability to adapt to changing reward distributions. In contrast, the SW-UCB and D-UCB algorithms, although more computationally intensive, exhibited superior performance in dynamic environments due to their recency bias.
Nevertheless, their performance is strongly influenced by the tuning of their key hyperparameters. LSTM networks performed well in stationary or slowly changing environments, yet their performance was limited in non-stationary conditions due to inherent assumptions of stationarity and challenges in tracking rapid changes. The results also showed that the choice of data handling methods, specifically the imputation method within the LSTM framework, can significantly impact performance. In conclusion, the choice of an algorithm for real-world applications should consider the specific requirements and constraints of the situation. Future research directions include enhancing the performance of UCB-1 in non-stationary environments, improving hyperparameter tuning for SW-UCB and D-UCB, adapting LSTM for non-stationary conditions, and exploring alternative data handling techniques for dealing with missing or inconsistent data. This study serves as a stepping stone towards a more nuanced understanding of different machine learning algorithms, underlining the importance of adaptability, computational efficiency, and performance trade-offs.
Ths diagram illustrates the progression of data from multiple cameras via the predictive unit, culminating in the facial recognition system.
- rofarsEnv.py - simulation environment
- agents.py - implementations for agents (LSTM and UCB agents by Jasper Bruin)
- example.py - example of how everything is put together
- UCBtest.py - testing and training script for the use of different UCB agents.
- RNNtest.py - testing and training script that uses historical traffic data from existing agents in agents.py
- data/train_test.txt (878,858 lines)
- the code supports Python 3.7+
pip install numpy==1.24.2 pandas==1.5.3 tqdm==4.64.1
- add your algorithm implementations to
agents.py - adapt
example.pyfor use in experiments
python example.py
- Jasper Bruin - Author of the LSTM and UCB agents
Cyril Hsu - [email protected]
Dr. Chrysa Papagianni - [email protected]
Jasper Bruin - [email protected]
GNU GENERAL PUBLIC LICENSE
Version 3, 29 June 2007
This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/.