Prediction of toxicity levels of chemical compounds is an important issue in Quantitative Structure-Activity Relationship (QSAR) modeling. Although toxicity prediction has achieved significant progress in recent times through deep learning, prediction accuracy levels obtained by even very recent methods are not yet very high. We propose a multimodal deep learning method using multiple heterogeneous neural network types and data representations. We represent chemical compounds by strings, images, and numerical features. We train fully connected, convolutional, and recurrent neural networks and their ensembles. Each data representation or neural network type has its own strengths and weaknesses. Our motivation is to obtain a collective performance that could go beyond individual performance of each data representation or each neural network type. On a standard toxicity benchmark, our proposed method obtains significantly better accuracy levels than that by the state-of-the-art toxicity prediction methods.

Character recognition being one of the most interesting and attractive areas of pattern recognition and artificial intelligence has got additional consideration during last decade due to its wide range of applications. It contributes immensely to the computerization process and enhancing the man-machine interaction in many applications. It is an art of detecting and recognizing the characters from input image and converting them into ASCII or other corresponding machine editable form. There are four main phases of Character Recognition -Data acquisition and Preprocessing, Segmentation, Feature extraction and Classification. Several research studies have been carried out for recognition of scripts like Chinese, Japanese, English, Devanagari, etc. but the research regarding Urdu Script is still immature due to cursive, variable and overlapping nature of Urdu characters and different writing styles. Research studies on printed Urdu characters have shown good recognition rate but the Handwritten Urdu Script Recognition is still an open and challenging area for researchers. This paper presents a review of Urdu handwritten character recognition methods with special reference to neural networks and includes information regarding the various operations that may be performed on the image for the recognition of Urdu characters. In literature, it has been found that B-Spline curves are not yet applied in combination with Neural Networks for Urdu script recognition. The current research work intends to use B-Splines curves for feature extraction with Neural Network as classifier and focuses on isolated characters in offline domain.

No abstract

No abstract

No abstract