Classification in Big Image Datasets Using Layered-SOM

Nakagawa, Akihiko; Kutics, Andrea

doi:10.1109/sitis.2017.33

Cited by 2 publications

(2 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unlike other SOM-based deep learning architectures such as Deep SOM, Deep Convolutional SOM, and Layered-SOM [22], [23], [24], the DendSOM proposed here is a shallow neural network that can be trained using unsupervised competitive learning algorithms. Hence, it can be applied to classification and CL tasks.…”

Section: Dendritic Self-organizing Mapmentioning

confidence: 99%

Dendritic Self-Organizing Maps for Continual Learning

Pinitas¹,

Chavlis²,

Poirazi³

2021

Preprint

View full text Add to dashboard Cite

Current deep learning architectures show remarkable performance when trained in large-scale, controlled datasets. However, the predictive ability of these architectures significantly decreases when learning new classes incrementally. This is due to their inclination to forget the knowledge acquired from previously seen data, a phenomenon termed catastrophic-forgetting. On the other hand, Self-Organizing Maps (SOMs) can model the input space utilizing constrained k-means and thus maintain past knowledge. Here, we propose a novel algorithm inspired by biological neurons, termed Dendritic-Self-Organizing Map (DendSOM). DendSOM consists of a single layer of SOMs, which extract patterns from specific regions of the input space accompanied by a set of hit matrices, one per SOM, which estimate the association between units and labels. The best-matching unit of an input pattern is selected using the maximum cosine similarity rule, while the point-wise mutual information is employed for class inference. DendSOM performs unsupervised feature extraction as it does not use labels for targeted updating of the weights. It outperforms classical SOMs and several state-of-the-art continual learning algorithms on benchmark datasets, such as the Split-MNIST and Split-CIFAR-10. We propose that the incorporation of neuronal properties in SOMs may help remedy catastrophic forgetting.

show abstract

Section: Dendritic Self-organizing Mapmentioning

confidence: 99%

Dendritic Self-Organizing Maps for Continual Learning

Pinitas¹,

Chavlis²,

Poirazi³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The SOM is a modified artificial neural network characterized by unsupervised training that can project high-dimensional information into a low-dimensional array [50]. In the SOM, input samples are compared based on the variable characteristics and mapped onto vectors referred to as neurons and are positioned closed to each other on a matrix referred as a map.…”

Section: Streamflowmentioning

confidence: 99%

Modelling Nitrate Reduction Strategies from Diffuse Sources in the Po River Basin

Malagó

Bouraoui

Pastori

et al. 2019

Water

View full text Add to dashboard Cite

Water contamination caused by the presence of excessive amounts of nitrate can be catastrophic for aquatic ecosystems and human health. Due to these high risks, a great deal of emphasis has been placed on finding effective measures to reduce nitrate concentrations in rivers and aquifers. In this study, we used the SWAT model based on grid-cells of 5 minutes of resolution for assessing the processes involved in nitrate loads generation and transport into aquifers and rivers and for providing basin management strategies of nitrate reduction. We applied the model in the Po River Basin (Italy), one of the most densely populated and highly agriculturally exploited area in the Mediterranean basin. The model was successfully calibrated and validated in eight monitoring stations along the Po River for the period 2000–2012. Simulated monthly streamflow and nitrate concentrations were in good agreement with observations, obtaining values of bias around ±25% in both calibration and validation. Among the tested scenarios of nitrogen reduction from agricultural sources, red clover cover crop after corn, coupled with a targeted reduction of mineral fertilizers and the limitation of nitrogen manure leads to a reduction of nitrate leaching and nitrogen emissions of around 37%.

show abstract

Classification in Big Image Datasets Using Layered-SOM

Cited by 2 publications

References 20 publications

Dendritic Self-Organizing Maps for Continual Learning

Dendritic Self-Organizing Maps for Continual Learning

Modelling Nitrate Reduction Strategies from Diffuse Sources in the Po River Basin

Contact Info

Product

Resources

About