Search

In this paper, we investigate Hebbian learning strategies applied to Convolutional Neural Network (CNN) training. We consider two unsupervised learning approaches, Hebbian Winner-Takes-All (HWTA), and Hebbian Principal Component Analysis (HPCA). The Hebbian learning rules are used to train the layers of a CNN in order to extract features that are then used for classification, without requiring backpropagation (backprop). Experimental comparisons are made with state-of-the-art unsupervised (but backprop-based) Variational Auto-Encoder (VAE) training. For completeness,we consider two supervised Hebbian learning variants (Supervised Hebbian Classifiers—SHC, and Contrastive Hebbian Learning—CHL), for training the final classification layer, which are compared to Stochastic Gradient Descent training. We also investigate hybrid learning methodologies, where some network layers are trained following the Hebbian approach, and others are trained by backprop. We tested our approaches on MNIST, CIFAR10, and CIFAR100 datasets. Our results suggest that Hebbian learning is generally suitable for training early feature extraction layers, or to retrain higher network layers in fewer training epochs than backprop. Moreover, our experiments show that Hebbian learning outperforms VAE training, with HPCA performing generally better than HWTA. [ABSTRACT FROM AUTHOR]

Seam carving is a content-aware algorithm for image resizing and tampering. This algorithm assigns an energy map to an image and removes the seams with low energy from the image. By doing this, seam carving makes it possible to reduce the image size and eliminate specific content from images. The detection of seam carving has lately been an important but challenging area of research. In past work, we had proposed a method that involved analyzing optimal patch types to recover seams and thus to detect seam-carved images. This method yielded highly accurate detection results. In this paper, we introduce an auto-associated Hopfield network to determine the optimal patch type for seam recovery. We use the Hebbian learning rule to choose, among candidate patch types, the one that most closely resembles the relevant target pattern. Experiments showed that the retrieval process usually converged within eight iterations and that the converged patterns improved the detection accuracy, e.g., with rates of 95.97 and 98.55% for 20 and 50% seam-carved images respectively. We also used this enhanced patch analysis method to identify the seam-carved regions of a tampered image. Its accuracy for the identification of tampered regions was higher than 70% for images with < 30% seam carving. [ABSTRACT FROM AUTHOR]

The detection of damage at an early stage that affects the supporting element of civil structures proves to be very significant to save invaluable human life and valuable possessions. In this research work, the severity of cracks in the supporting column is assessed using a new technique. This piece of research study uses the soft computing method of fuzzy cognitive map (FCM) to model the domain experts' knowledge and the knowledge assimilated through relevant literature to grade the severity of cracks in supporting column. The FCM grading model is further improved by using the Hebbian learning algorithms. The presented work demonstrates the classification and prediction capabilities of FCM for the respective structural health monitoring application, using two well-known and efficient FCM learning approaches viz. nonlinear Hebbian learning (NHL) and data-driven nonlinear Hebbian learning (DD-NHL). The proposed crack severity grading model classifies the cracks in supporting column into three categories, namely fine crack, moderate crack and severe crack. The proposed model uses DD-NHL algorithm. DD-NHL is trained with 70 records and tested with 30 records and gives an overall classification accuracy of 96 %. The obtained results are better compared to other popular machine learning-based classifiers. The proposed method helps even the non-experts to find the possible causes of crack and reports them to structural engineers, to start maintenance in an appropriate stage, using various crack control techniques. Also, a software tool for crack categorization was developed based on the FCM method and its learning capabilities. Thus, it is easier for the users/civil engineers to use this software to make decisions in civil engineering domain and improve their knowledge about the health of the structure. [ABSTRACT FROM AUTHOR]

Coconut is one of the major perennial food crops that has a long development phase of 44 months. The climatic and seasonal variations affect all stages of coconut's long development cycle. Besides, the soil composition also plays a vital role in deciding the coconut yield behavior. The present study is focused on categorizing the coconut production level for the given set of agro-climatic conditions using the methodology of fuzzy cognitive map (FCM) enhanced by its learning capabilities. Additionally, an attempt is made to study the impact of climatic variations and weather parameters on the coconut yield behavior using the reasoning capabilities of FCM. Real coconut field data of different seasons for the period from 2009 to 2013 of Kerala state's Malabar region were used for training and evaluation of the FCM. The present work demonstrates the classification and prediction capabilities of FCM for the described precision agriculture application, with the two most known and efficient FCM learning approaches, viz., nonlinear Hebbian (NHL) and data-driven nonlinear Hebbian (DDNHL). The DDNHL-FCM offers an overall classification accuracy of 96 %. The various case studies furnished in the paper demonstrate the power of NHL-FCM in effectively reasoning new knowledge pertaining to the presented precision agriculture application. [ABSTRACT FROM AUTHOR]

In this paper, a target tracking controller based on spiking neural network is proposed for autonomous robots. This controller encodes the preprocessed environmental and target information provided by CCD cameras, encoders and ultrasonic sensors into spike trains, which are integrated by a three-layer spiking neural network (SNN). The outputs of SNN are generated based on the competition between the forward/backward neuron pair corresponding to each motor, with the weights evolved by the Hebbian learning. The application to target tracking of a mobile robot in unknown environment verifies the validity of the proposed controller. [ABSTRACT FROM AUTHOR]

It is now widely accepted that memristive devices are promising candidates for the emulation of the behavior of biological synapses in neuromorphic systems. This is mainly because of the fact that like the strength of synapse, memristance of the memristive device can be tuned actively for example by the application of voltage or current. In addition, it is also possible to fabricate high density of memristive devices through the nano-crossbar structures. In this paper, we will show that there are some problems associated with memristive devices, which are playing the role of biological synapses. For example, we show that the variation rate of the memristance depends completely on the initial state of the device, and therefore, it can change significantly with time during the learning phase. This phenomenon can degrade the performance of learning methods like spike timing-dependent plasticity and cause the corresponding neuromorphic systems to become unstable. We also illustrate that using two serially connected memristive devices with different polarities as a synapse can somewhat fix the aforementioned problem. [ABSTRACT FROM AUTHOR]

Simulated networks of spiking leaky integrators are used to categorise and for Information Retrieval (IR). Neurons in the network are sparsely connected, learn using Hebbian learning rules, and are simulated in discrete time steps. Our earlier work has used these models to simulate human concept formation and usage, but we were interested in the model’s applicability to real world problems, so we have done experiments on categorisation and IR. The results of the system show that congresspeople are correctly categorised 89% of the time. The IR systems have 40% average precision on the Time collection, and 28% on the Cranfield 1,400. All scores are comparable to the state of the art results on these tasks. [ABSTRACT FROM AUTHOR]