Physics-Constrained Comprehensive Optical Neural Networks
Authors: Yanbing Liu, JIANWEI QIN, Yan Liu, Xi Yue, Xun Liu, Guoqing Wang, Tianyu Li, Ye, Wei Li
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This work conducts extensive experiments to investigate systematic errors in the optical physical system within the context of image classification tasks. Through our investigation, two quantifiable errors light source instability and exposure time mismatches significantly impact the prediction performance of ONN. To address these systematic errors, a physics-constrained ONN learning framework is constructed... In our experiments, the proposed method achieved a test classification accuracy of 96.5% on the MNIST dataset, a substantial improvement over the 61.6% achieved with the original ONN. |
| Researcher Affiliation | Academia | 1 Beijing University of Posts and Telecommunications, Beijing,100876,China 2 Shanghai Jiao Tong University, Shanghai 200240, China 3 Beijing Institute of Space Mechanics and Electricity,Beijing,100094,China 4 University of Electronic Science and Technology of China, Chengdu, 611731, China |
| Pseudocode | No | The paper describes the training process in section 3.3 using numbered steps (i)-(iv), but it is presented as descriptive text rather than a formally structured pseudocode or algorithm block. |
| Open Source Code | Yes | The paper's self-assessment in the NeurIPS checklist states 'Answer: [Yes] Justification: Supplemental material and Section4' for the question 'Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material?' implying code availability via supplementary materials. |
| Open Datasets | Yes | The MNIST (Modified National Institute of Standards and Technology)[50, 51] dataset consists of 70,000 28x28 pixel grayscale images of handwritten digits... The Quick Draw16 dataset, a subset of Google s Quick, Draw! project... The Fashion MNIST dataset[52], provided by Zalando, contains 70,000 28x28 pixel grayscale images... |
| Dataset Splits | No | During training, we used the entire training set. To evaluate accuracy, we randomly selected 1,000 images from the test sets of each of the three datasets to assess experimental accuracy. The paper mentions training and test sets but does not specify a separate validation set or provide distinct train/validation splits. |
| Hardware Specification | No | The paper describes the optical experimental setup (Fig 3(a)) but does not specify any particular GPU or CPU models, memory, or other computational hardware used for training the neural networks. While the NeurIPS checklist mentions supplementary material for compute resources, this information is not present in the main paper. |
| Software Dependencies | No | The paper describes the methods and processes but does not list specific software dependencies with their version numbers (e.g., Python, PyTorch, specific libraries). |
| Experiment Setup | Yes | For the instability of laser intensity fjit, we used a CCD to measure the overall grayscale values within a fixed area... A well-designed loss function is used to increase the gap between the maximum and second maximum intensity values within the classification area... When WGap is set to 10, the experimental accuracy is optimal. In this setting, the network still maintains good fitting ability, and the instability of laser intensity is also compensated. This approach ensures that despite fluctuations in laser output, the performance of the optical neural network remains robust, providing reliable and precise experimental outcomes. |