Stuart Gibson

T. Tula, G. Möller, J. Quintanilla et al.

Machine learning (ML) methods have proved to be a very successful tool in physical sciences, especially when applied to experimental data analysis. Artificial intelligence is particularly good at recognizing patterns in high dimensional data, where it usually outperforms humans. Here we applied a simple ML tool called principal component analysis (PCA) to study data from muon spectroscopy. The measured quantity from this experiment is an asymmetry function, which holds the information about the average intrinsic magnetic field of the sample. A change in the asymmetry function might indicate a phase transition; however, these changes can be very subtle, and existing methods of analyzing the data require knowledge about the specific physics of the material. PCA is an unsupervised ML tool, which means that no assumption about the input data is required, yet we found that it still can be successfully applied to asymmetry curves, and the indications of phase transitions can be recovered. The method was applied to a range of magnetic materials with different underlying physics. We discovered that performing PCA on all those materials simultaneously can have a positive effect on the clarity of phase transition indicators and can also improve the detection of the most important variations of asymmetry functions. For this joint PCA we introduce a simple way to track the contributions from different materials for a more meaningful analysis.

Fangliang Bai, Jinchao Liu, Xiaojuan Liu et al.

Recent work showed neural-network-based approaches to reconstructing images from compressively sensed measurements offer significant improvements in accuracy and signal compression. Such methods can dramatically boost the capability of computational imaging hardware. However, to date, there have been two major drawbacks: (1) the high-precision real-valued sensing patterns proposed in the majority of existing works can prove problematic when used with computational imaging hardware such as a digital micromirror sampling device and (2) the network structures for image reconstruction involve intensive computation, which is also not suitable for hardware deployment. To address these problems, we propose a novel hardware-friendly solution based on mixed-weights neural networks for computational imaging. In particular, learned binary-weight sensing patterns are tailored to the sampling device. Moreover, we proposed a recursive network structure for low-resolution image sampling and high-resolution reconstruction scheme. It reduces both the required number of measurements and reconstruction computation by operating convolution on small intermediate feature maps. The recursive structure further reduced the model size, making the network more computationally efficient when deployed with the hardware. Our method has been validated on benchmark datasets and achieved the state of the art reconstruction accuracy. We tested our proposed network in conjunction with a proof-of-concept hardware setup.

Jinchao Liu, Stuart J. Gibson, Margarita Osadchy

Deep Learning shows very good performance when trained on large labeled data sets. The problem of training a deep net on a few or one sample per class requires a different learning approach which can generalize to unseen classes using only a few representatives of these classes. This problem has previously been approached by meta-learning. Here we propose a novel meta-learner which shows state-of-the-art performance on common benchmarks for one/few shot classification. Our model features three novel components: First is a feed-forward embedding that takes random class support samples (after a customary CNN embedding) and transfers them to a better class representation in terms of a classification problem. Second is a novel attention mechanism, inspired by competitive learning, which causes class representatives to compete with each other to become a temporary class prototype with respect to the query point. This mechanism allows switching between representatives depending on the position of the query point. Once a prototype is chosen for each class, the predicated label is computed using a simple attention mechanism over prototypes of all considered classes. The third feature is the ability of our meta-learner to incorporate deeper CNN embedding, enabling larger capacity. Finally, to ease the training procedure and reduce overfitting, we averages the top $t$ models (evaluated on the validation) over the optimization trajectory. We show that this approach can be viewed as an approximation to an ensemble, which saves the factor of $t$ in training and test times and the factor of of $t$ in the storage of the final model.

Jinchao Liu, Stuart J. Gibson, James Mills et al.

Convolutional neural networks (CNN) have been shown to provide a good solution for classification problems that utilize data obtained from vibrational spectroscopy. Moreover, CNNs are capable of identification from noisy spectra without the need for additional preprocessing. However, their application in practical spectroscopy is limited due to two shortcomings. The effectiveness of the classification using CNNs drops rapidly when only a small number of spectra per substance are available for training (which is a typical situation in real applications). Additionally, to accommodate new, previously unseen substance classes, the network must be retrained which is computationally intensive. Here we address these issues by reformulating a multi-class classification problem with a large number of classes, but a small number of samples per class, to a binary classification problem with sufficient data available for representation learning. Namely, we define the learning task as identifying pairs of inputs as belonging to the same or different classes. We achieve this using a Siamese convolutional neural network. A novel sampling strategy is proposed to address the imbalance problem in training the Siamese Network. The trained network can effectively classify samples of unseen substance classes using just a single reference sample (termed as one-shot learning in the machine learning community). Our results demonstrate better accuracy than other practical systems to date, while allowing effortless updates of the system's database with novel substance classes.

Fangliang Bai, Manuel J. Marques, Stuart J. Gibson

In this paper we present a new method for cystoid macular edema (CME) segmentation in retinal Optical Coherence Tomography (OCT) images, using a fully convolutional neural network (FCN) and a fully connected conditional random fields (dense CRFs). As a first step, the framework trains the FCN model to extract features from retinal layers in OCT images, which exhibit CME, and then segments CME regions using the trained model. Thereafter, dense CRFs are used to refine the segmentation according to the edema appearance. We have trained and tested the framework with OCT images from 10 patients with diabetic macular edema (DME). Our experimental results show that fluid and concrete macular edema areas were segmented with good adherence to boundaries. A segmentation accuracy of $0.61\pm 0.21$ (Dice coefficient) was achieved, with respect to the ground truth, which compares favourably with the previous state-of-the-art that used a kernel regression based method ($0.51\pm 0.34$). Our approach is versatile and we believe it can be easily adapted to detect other macular defects.

Jinchao Liu, Margarita Osadchy, Lorna Ashton et al.

Machine learning methods have found many applications in Raman spectroscopy, especially for the identification of chemical species. However, almost all of these methods require non-trivial preprocessing such as baseline correction and/or PCA as an essential step. Here we describe our unified solution for the identification of chemical species in which a convolutional neural network is trained to automatically identify substances according to their Raman spectrum without the need of ad-hoc preprocessing steps. We evaluated our approach using the RRUFF spectral database, comprising mineral sample data. Superior classification performance is demonstrated compared with other frequently used machine learning algorithms including the popular support vector machine.

Mor Ohana, Orr Dunkelman, Stuart Gibson et al.

One of the main challenges faced by Biometric-based authentication systems is the need to offer secure authentication while maintaining the privacy of the biometric data. Previous solutions, such as Secure Sketch and Fuzzy Extractors, rely on assumptions that cannot be guaranteed in practice, and often affect the authentication accuracy. In this paper, we introduce HoneyFaces: the concept of adding a large set of synthetic faces (indistinguishable from real) into the biometric "password file". This password inflation protects the privacy of users and increases the security of the system without affecting the accuracy of the authentication. In particular, privacy for the real users is provided by "hiding" them among a large number of fake users (as the distributions of synthetic and real faces are equal). In addition to maintaining the authentication accuracy, and thus not affecting the security of the authentication process, HoneyFaces offer several security improvements: increased exfiltration hardness, improved leakage detection, and the ability to use a Two-server setting like in HoneyWords. Finally, HoneyFaces can be combined with other security and privacy mechanisms for biometric data. We implemented the HoneyFaces system and tested it with a password file composed of 270 real users. The "password file" was then inflated to accommodate up to $2^{36.5}$ users (resulting in a 56.6 TB "password file"). At the same time, the inclusion of additional faces does not affect the true acceptance rate or false acceptance rate which were 93.33\% and 0.01\%, respectively.