Reservoir Computing with Untrained Convolutional Neural Networks for Image Recognition

Boson sampling has demonstrated quantum advantage yet struggled to find real world practical applications. We show these processors generate the complex dynamics necessary to power quantum reservoir computing and utilize it for various image recognition and classification tasks.

With the rapid development of science and technology, ultrasound has been paid more and more attention by people, and it is widely used in engineering, diagnosis, and detection. In this paper, an ultrasonic image recognition method based on immune algorithm is proposed for ultrasonic images, and its method is applied to medical ultrasound liver image recognition. Firstly, this paper grays out the ultrasound liver image and selects the region of interest of the image. Secondly, it extracts the feature based on spatial gray matrix independent matrix, spatial frequency decomposition, and fractal features. Then, the immune algorithm is used to classify and identify the normal liver, liver cirrhosis, and liver cancer ultrasound images. Finally, based on the deficiency of the immune algorithm, it is combined with the support vector machine to form an optimized immune algorithm, which improves the performance of ultrasonic liver image classification and recognition. The simulation shows that this paper can effectively classify the normal liver, liver cirrhosis, and liver cancer ultrasound images. Compared with the traditional immune algorithm, this paper combines the immune algorithm with the support vector machine, and the optimized immune algorithm can effectively improve the performance of ultrasonic liver image classification and recognition.

In order to improve the accuracy and speed of image recognition, an improved image recognition algorithm is designed based on genetic algorithm and BP neural network as results of the traditional BP neural network has problems such as structural parameter uncertainty, low convergence rate and local minimum value. Firstly, the color and texture feature of the image are extracted. BP neural network structure parameters are optimized by using BP neural network based on the genetic algorithm. The image recognition process is presented and the image recognition result is integrated according to the evidence theory to obtain the complete image information. The simulation results show that the algorithm has high recognition rate and convergence speed. Under the condition of a small number of training samples, BP neural network still has better generalization ability.

This research aims to study the deep learning applications in image recognition and classification tasks. The advantages and limitations of deep learning are explored by analysing existing deep learning algorithms and their applications on tasks such as using images to do classification, image segmentation, and target detection. In the experimental part, this paper evaluates the performance of deep learning in image recognition by using classical neural network models, including training and testing models on a large size of image datasets. The results show that neural network has good classification and detection capabilities in image recognition tasks and also achieves good results for image segmentation tasks. However, the model training process of shallow neural network models is time-consuming and performs poorly for small-scale datasets. Based on these situations, this paper proposes some optimization strategies to achieve high performance and efficiency of neural networks in image recognition and analyze the performance differences between different strategies .

Nowadays, occupational safety incidents have caused a large number of casualties during operation or maintenance in the factories. For example, it could happen during the compression process in the plastic injection molding industry that when a technician is in the middle of two molds to dismantle them and another personnel outside the work area does not pay special attention and accidentally presses the start button, the technical working between the molds may be instantly clamped by the molds, causing serious casualties. Also, we may easily bring about another kind of example in heavy industry: during the operation of a punching machine, the operator could be pressed by the punching die due to a momentary lapse of concentration, resulting in broken fingers or injuries to the arm, elbow, or wrist. Therefore, this study aims to apply image and speech recognition to the robotic arm so as to control the robotic arm by voice, or independently perform image recognition through verbal commands. In this way, the operator does not need to approach the machine. For instance, the voice-controlled robotic arm can perform image recognition to identify the type of object, and then the arm will pick it up and place it in the designated location. Additionally, the image recognition scheme can be enabled separately during the calibration stage to take pictures of the object to be recognized in the subsequent production process. In other words, it helps operators avoid accessing high-risk areas by employing verbal control for tasks that require human intervention. The simulation and actual test have evidenced that image recognition can be enabled independently by using speech recognition, and the robotic arm can be controlled directly by verbal commands to achieve remote control.

In this paper, we report our studies on integrated active pixel sensor (APS) array and memristor crossbar neural network to perform image learning and recognition in an unsupervised fashion. APS modules encode light intensity/gray value of grayscale images into APS sensing current feeding into N <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ×M crossbar array. The memristor is used as a synapse and can be trained through an adaption of spike timing dependent plasticity (STDP). After training, different images are stored into different post-synaptic neuron dendrites. In the image recognition stage, a simple pulse counter circuit was used to check the matched image. System level simulations show that the network can store grayscale image correctly and perform image recognition in a simple and efficient way.

Fruit classification has become increasingly important in a wide range of industrial and consumer-oriented applications. Automated fruit classification systems can significantly enhance efficiency by accurately identifying fruit varieties and supporting informed decisions. In this research, we propose a fast, accurate, and robust fruit classification approach leveraging Deep Learning (DL) techniques. The proposed approach is a fine-tuned, pretrained Visual Geometry Group-19 (VGG-19) convolutional neural network model, known for its depth and superior performance in image recognition and classification tasks. Unlike existing approaches, our proposed approach addresses the unique visual challenges of fruit images, including inter-class similarity, intra-class variability, occlusions, background clutter, and varying illumination conditions, which enhances generalization across both controlled and real-world datasets. This approach is rigorously evaluated using two distinct datasets obtained from Kaggle. Dataset 1 comprises high-resolution images captured under controlled conditions, while Dataset 2, derived from the Kaggle 360 Fruits dataset, contains diverse real-world images with varying backgrounds, lighting conditions, and occlusions. Experimental results demonstrate that the proposed model achieves exceptional classification accuracy, recording 99.65% on Dataset 1 and 97.98% on Dataset 2. The above findings underline the accuracy and stability of the fine-tuned VGG-19 model in processing clean and complicated images and pose a viable and scalable real-time fruit classification approach.

With the growing use of artificial neural networks (ANNs) in temporal data processing tasks, the cost of training for complex ANNs is an escalating concern. Physical reservoir computing (RC), a variation of recurrent neural networks (RNNs), obviate the need for most data intensive matrix vector multiplication in the recurrent layer by evolving the RC’s internal states with the inherent nonlinear dynamics and short-term memory. In this study, we show that magnetic skyrmion confined in a fixed geometry forming the soft layer of a magnetic tunnel junction (MTJ) can work as a RC and perform autonomous long-term prediction of temporal data. Our proposed skyrmion reservoir allows for manipulation of spin dynamics with ultra energy efficient voltage controlled magnetic anisotropy modulation (VCMA) method. Furthermore, the boundary effect on the skyrmion from the geometric edges provides necessary consistency property of the reservoir. We employ our proposed reservoir for the modeling and prediction of the chaotic time series such as Mackey-Glass and dynamic time-series data, such as household building energy loads. For autonomous run, the predicted output is fed to the input of the reservoir. By comparing our spintronic physical RC approach with energy load forecasting algorithms, such as LSTMs and RNNs, we conclude that the proposed framework presents good performance in achieving high predictions accuracy, while also requiring low memory and energy both of which are at a premium in hardware resource and power constrained edge applications. Further, the proposed approach is shown to require very small training datasets and at the same time being at least 16× energy efficient compared to the sequence to sequence LSTM for accurate household load predictions. Higher endurance, fast processing and well-established technology (i.e., MTJ) to integrate with CMOS technology makes such spintronic reservoir attractive over other emergent memory device-based reservoirs.

We present a complex network of interconnected optical structures for information processing. This network is an implementation of reservoir computing, a novel method in the field of machine learning. Reservoir computing can be used for example in classification problems such as speech and image recognition, or for the generation of arbitrary patterns, tasks which are usually very hard to generalize. A nanophotonic reservoir can be constructed to perform optical signal processing. Previously, simulations demonstrated that a reservoir consisting of Semiconductor Optical Amplifiers (SOA) can outperform traditional software-based reservoirs for a speech task. Here we propose a network of coupled photonic crystal cavities. Because of the resonating behaviour, a lot of power is stored in the cavity, which gives rise to interesting nonlinear effects. Simulations are done using a novel software tool developed at Ghent University, called Caphe. We train this network of coupled resonators to generate a periodic pattern using a technique called FORCE. It is shown that photonic reservoirs can outperform classical software-based reservoirs on a pattern generation task.

As artificial intelligence continues to push into real-time, edge-based and resource-constrained environments, there is an urgent need for novel, hardware-efficient computational models. In this study, we present and validate a neuromorphic computing architecture based on resonant-tunnelling diodes (RTDs), which exhibit the nonlinear characteristics ideal for physical reservoir computing (RC). We theoretically formulate and numerically implement an RTD-based RC system and demonstrate its effectiveness on two image recognition benchmarks: handwritten digit classification and object recognition using the Fruit-360 dataset. Our results show that this circuit-level architecture delivers promising performance while adhering to the principles of next-generation RC, eliminating random connectivity in favour of a deterministic nonlinear transformation of input signals.

Real-world applications such as first-person video activity recognition require intelligent edge devices. However, size, weight, and power constraints of the embedded platforms cannot support resource intensive state-of-the-art algorithms. Machine learning lite algorithms, such as reservoir computing, with shallow 3-layer networks are computationally frugal as only the output layer is trained. By reducing network depth and plasticity, reservoir computing minimizes computational power and complexity, making the algorithms optimal for edge devices. However, as a trade-off for their frugal nature, reservoir computing sacrifices computational power compared to state-of-the-art methods. A good compromise between reservoir computing and fully supervised networks are the proposed deep-LSM networks. The deep-LSM is a deep spiking neural network which captures dynamic information over multiple time-scales with a combination of randomly connected layers and unsupervised layers. The deep-LSM processes the captured dynamic information through an attention modulated readout layer to perform classification. We demonstrate that the deep-LSM achieves an average of 84.78% accuracy on the DogCentric video activity recognition task, beating state-of-the-art. The deep-LSM also shows up to 91.13% memory savings and up to 91.55% reduction in synaptic operations when compared to similar recurrent neural network models. Based on these results we claim that the deep-LSM is capable of overcoming limitations of traditional reservoir computing, while maintaining the low computational cost associated with reservoir computing.

Reservoir computing as one of the artificial neural networks can process input signals in various ways, thereby showing strength in modeling data that changes over time. The reservoir is utilized in various fields because it is particularly energy efficient in learning and can exhibit powerful performance with relatively few trainings cost. This study emphasizes the significant advantages of the hafnium zirconium oxide (HZO) film in reservoir applications by controlling the depolarization field. The decay time of HZO-based ferroelectric memory devices is investigated, focusing on the impact of both ferroelectric layer thickness and interlayer (IL) thickness on physical reservoir computing system. Devices with HZO film thicknesses of 10, 15, and 20 nm were fabricated and characterized. Among these, the 15 nm HZO film demonstrated optimal thickness, exhibiting excellent ferroelectric properties, including enhanced orthorhombic phase (o-phase) formation and reliable short-term memory characteristics. When the optimized device for decay time is integrated into a reservoir computing system, it achieved a remarkable average accuracy of 93.42% in image recognition tasks, emphasizing its capability for high-precision computations.

Photonic reservoir computing has been intensively investigated to solve machine learning tasks effectively. A simple learning procedure of output weights is used for reservoir computing. However, the lack of training of input-node and inter-node connection weights limits the performance of reservoir computing. The use of multiple reservoirs can be a solution to overcome this limitation of reservoir computing. In this study, we investigate parallel and deep configurations of delay-based all-optical reservoir computing using semiconductor lasers with optical feedback by combining multiple reservoirs to improve the performance of reservoir computing. Furthermore, we propose a hybrid configuration to maximize the benefits of parallel and deep reservoirs. We perform the chaotic time-series prediction task, nonlinear channel equalization task, and memory capacity measurement. Then, we compare the performance of single, parallel, deep, and hybrid reservoir configurations. We find that deep reservoirs are suitable for a chaotic time-series prediction task, whereas parallel reservoirs are suitable for a nonlinear channel equalization task. Hybrid reservoirs outperform other configurations for all three tasks. We further optimize the number of reservoirs for each reservoir configuration. Multiple reservoirs show great potential for the improvement of reservoir computing, which in turn can be applied for high-performance edge computing.

Analysis and fully memristor-based reservoir computing for temporal data classification

Nowadays mobile devices such as smartphones are widely used all over the world. Moreover, the performance of image recognition has dramatically increased by deep learning technologies. From these backgrounds, we think that the following scenario of information services could be realized in the near future: users take a photo and send it to a server, who recognizes the location in the photo and returns the users some information about the recognized location. However, this kind of client-server-based image recognition can cause a privacy issue because image recognition results are sometimes privacy sensitive. To tackle the privacy issue, in this paper, we propose a novel framework for privacy-preserving image recognition in which the server cannot uniquely identify the recognition result but users can do so. An overview of the proposed framework is as follows: First users extract a visual feature from their taken photo and transform it so that the server cannot uniquely identify the recognition result. Then users send the transformed feature to the server, who returns a candidate set of recognition results to the users. Finally, the users compare the candidates and the original visual feature for obtaining the final result. Our experimental results demonstrate the effectiveness of the proposed framework.