Event recognition in soccer is a challenging task due to the complexity of the game, the number of players and the subtle differences among different actions. In this work, we introduce the SoccER dataset, which is generated by an automatic system b…

We defined a basical criterium to select a transient event from a temporal series (i.e. T90 as we made in (Feroci et al. 1999; Lazzarotto 2001)

Model compression is a common way to deal with latency-critical or memory-intensive scenarios. Existing model compression methods for BERT need to be re-trained on a large corpus to reserve its original performance, which is inapplicable in low-reso…

Standard NER datasets such as English news contain a limited number of unique entities, with many of them occurring in both the train and the test sets. State-of-the-art models may thus memorize observed names, rather than learning to identify entit…

We have shown some issues of the Advantage Actor Critic (A2C) algorithm that arise from the scalarization of a multi-objective optimization problem. We proposed a way to avoid gradient overlapping (NOG) and a way to control the entropy (TE) of sampl…

The experiments have shown that detection of concept evolution utilising deep neural network activations via streaming detection methods is a viable approach. This was proven using two separated types of classes (transport and animals) from the CIFA…

In this paper, we have presented the DRL scheme for the resource management in the BFL system. We have first described the BFL system in which the blockchain network is modeled as an M/M/1 queue. Then, we have formulated the stochastic optimization …

Super-resolution is a tool of interest in many applications namely, remote sensing, medical applications, entertainment, where these techniques are intended to improve the user experience. In all these applications, besides reconstruction performanc…

In this paper, we made a discovery that only a few labelled points is needed for existing point cloud encoder networks to produce very competitive performance for the point cloud segmentation task. We provide analysis from a statistical point of vie…

Computer vision methods have great promise as a means of shielding farmers from the negative effects of climate change due to their ability to flexibly map satellite data to information on yields, forage quality, or other measures collected from the…

The main objective of this work is to draw attention to the problem of CL for sequential data processing by introducing GIM, a recurrent CL architecture inspired by the Progressive networks [rusu_progressive_2016]. Our benchmarks show that GIM is ab…

We have provided a series of analyses that demonstrate together that non-isomorphism is not—as previously assumed—primarily and solely a result of typological differences between languages, but due in large part to degenerate vector spaces and discr…

We presented GazSelfAttn, a novel approach for gazetteer embeddings that uses self-attention and match span positions. Evaluation results of GazSelfAttn show improvement compared to competitive baselines and state-of-the-art models on multiple datas…

In this paper, we propose a novel model called DynaBERT which can run at different widths and depths. Experiments on various tasks show that under the same efficiency constraint, sub-networks extracted from the proposed DynaBERT consistently achieve…

In this paper, the use of neural network (NN) and O-D travel time matrix for travel time prediction was explored. Two neural network models namely multi-layer perceptron (MLP) and long short-term model (LSTM) were built and compared based on predict…

subfiles/06_conclusion

This paper proposes pretrained deep learning models for determining severity status of T.Absoluta tomato damages plants. For the accomplishment of this work, we used images containing health and T.Absoluta infested tomato plant images collected fro…