Papers by Narendra Chaudhari
We introduce a method for construction ofa Turing machine using binary, hard-limiter neurons with... more We introduce a method for construction ofa Turing machine using binary, hard-limiter neurons with integer weights and thresholds. We identify the problem of potentially infinite fan-in needed fur read units. We give two approaches to tackle this problem. The first approach organizes the neural read units in the form of a pipeline. The second approach organizes the read units in tree-structure. We identify trade-off in time performance and design complexity for these two approaches.
Efficient Incremental Model for Learning Context-Free Grammars from Positive Structural Examples
This paper describes a formalization based on tree automata for incremental learning of context-f... more This paper describes a formalization based on tree automata for incremental learning of context-free grammars from positive samples of their structural descriptions. A structural description of a context-free grammar is a derivation tree of the grammar in which labels are removed. The tree automata based learning in this paradigm is early introduced by Sakakibara in 1992, however his scheme assumes that all training examples are available to the learning algorithm at the beginning (i.e., it cannot be employed as an online learning) and also it doesn’t optimize the storage requirements as well. Our model has several desirable features that runs in O(n 3) time in the sum of the sizes of the input examples, obtains O(n) storage space saving, achieves good incremental behavior by updating a guess incrementally and infers a grammar from positive-only examples efficiently. Several examples and experimental results are given to illustrate the scheme and its efficient execution.
A multiclass classifier using Genetic Programming
... A Multiclass Classifier Using Genetic Programming Narendra S. Chaudhari Anuradha PurohitAruna... more ... A Multiclass Classifier Using Genetic Programming Narendra S. Chaudhari Anuradha PurohitAruna Tiwari ... A. Datasets 1) WBC : The breast cancer database was obtained from the University of Wisconsin Hospitals, Madison from Dr. William H. Wolberg. ...
Classification Automaton and Its Construction Using Learning
A method of regular grammar inference in computational learning for classification problems is pr... more A method of regular grammar inference in computational learning for classification problems is presented. We classify the strings by generating multiple subclasses. A construction algorithm is presented for the classification automata from positive classified examples. We prove the correctness of the algorithm, and suggest some possible extensions.
IEEE Transactions on Neural Networks, 2009
Conventional recurrent neural networks (RNNs) have difficulties in learning long-term dependencie... more Conventional recurrent neural networks (RNNs) have difficulties in learning long-term dependencies. To tackle this problem, we propose an architecture called segmented-memory recurrent neural network (SMRNN). A symbolic sequence is broken into segments and then presented as inputs to the SMRNN one symbol per cycle. The SMRNN uses separate internal states to store symbol-level context, as well as segment-level context. The symbol-level context is updated for each symbol presented for input. The segment-level context is updated after each segment. The SMRNN is trained using an extended real-time recurrent learning algorithm. We test the performance of SMRNN on the information latching problem, the "two-sequence problem" and the problem of protein secondary structure (PSS) prediction. Our implementation results indicate that SMRNN performs better on long-term dependency problems than conventional RNNs. Besides, we also theoretically analyze how the segmented memory of SMRNN helps learning long-term temporal dependencies and study the impact of the segment length.
Capturing Long-Term Dependencies for Protein Secondary Structure Prediction
Bidirectional recurrent neural network (BRNN) is a noncausal system that captures both upstream a... more Bidirectional recurrent neural network (BRNN) is a noncausal system that captures both upstream and downstream information for protein secondary structure prediction. Due to the problem of vanishing gradients, the BRNN can not learn remote information efficiently. To limit this problem, we propose segmented memory recurrent neural network (SMRNN) and obtain a bidirectional segmented-memory recurrent neural network (BSMRNN) by replacing the standard RNNs in BRNN with SMRNNs. Our experiment with BSMRNN for protein secondary structure prediction on the RS126 set indicates improvement in the prediction accuracy.
Neurocomputing, 2006
For binary neural networks (BNNs), constructive covering frameworks have been investigated recent... more For binary neural networks (BNNs), constructive covering frameworks have been investigated recently. While these frameworks are fast, they have limitations of generalization and accurate classification for learning from limited number of samples. In this paper, we propose modified constructive-covering algorithm (MCCA), which consists of two processes: generalization process and modification process. Errors introduced in the generalization process are revised in the modification process by adding modification neurons. In our approach, we visualize hidden neurons in terms of hypershperes. The learning process is the geometrical expansion process of these hypershperes. Through our experimental work in Section 5, we conclude that, MCCA is not sensitive to the order in which the input sequence is given. In addition, MCCA results in simple neural network structures by less training time. r
Eurasip Journal on Advances in Signal Processing, 2005
We present an artificial neural-network-(NN-) based smart interface framework for sensors operati... more We present an artificial neural-network-(NN-) based smart interface framework for sensors operating in harsh environments. The NN-based sensor can automatically compensate for the nonlinear response characteristics and its nonlinear dependency on the environmental parameters, with high accuracy. To show the potential of the proposed NN-based framework, we provide results of a smart capacitive pressure sensor (CPS) operating in a wide temperature range of 0 to 250 • C. Through simulated experiments, we have shown that the NN-based CPS model is capable of providing pressure readout with a maximum full-scale (FS) error of only ±1.0% over this temperature range. A novel scheme for estimating the ambient temperature from the sensor characteristics itself is proposed. For this purpose, a second NN is utilized to estimate the ambient temperature accurately from the knowledge of the offset capacitance of the CPS. A microcontroller-unit-(MCU-) based implementation scheme is also provided.
A constructive unsupervised learning algorithm for clustering binary patterns
We propose a constructive unsupervised learning algorithm (CULA) for Boolean neural networks base... more We propose a constructive unsupervised learning algorithm (CULA) for Boolean neural networks based on geometrical expansion. CULA constructs two-layered (input and output layer) neural networks. We visualize output neurons in terms of hyperspheres. CULA results in fast learning because it determines whether to add a new coming vertex to a neuron by its geometrical location, not by iterant computation. We illustrate CULA by using 101 instances in zoo database of Richard Forsyth, and compare our unsupervised clustering with clustering by biological experts given in the zoo database.
XML query algebra operators, and strategies for their implementation
Page 1. _____ 0-7803-8560-8/04/$20.00©2004IEEE XML QUERY ALGEBRA OPERATORS, AND STRATEGIES FOR TH... more Page 1. _____ 0-7803-8560-8/04/$20.00©2004IEEE XML QUERY ALGEBRA OPERATORS, AND STRATEGIES FOR THEIR IMPLEMENTATION Jacob Abraham, Narendra S. Chaudhari1and Edmond C. Prakash2 ...
Time series prediction using principal feature analysis
Abstract We give the formulation for time series prediction using principal feature analysis (PFA... more Abstract We give the formulation for time series prediction using principal feature analysis (PFA). PFA is a method introduced by Ira Cohen, Qi Tian et al. in 2002 for feature subset selection problem. PFA involves k-means formulation on significant principal components, ...
In this paper, we present a method for extencling the Expand and Truncate Leuming (ETL) [3] techn... more In this paper, we present a method for extencling the Expand and Truncate Leuming (ETL) [3] technique for multi-class output. Kim and Park [3] have given two hints for solving the same problem, but their hints may result in a large sued neural net. Here, we propose another method for the same problem, which is expected LO result in a smaller neural net. Our method is based on yet other known technique introduced by Gray and Michel [Z], called Boolean like training algorithm (BLTA).
Improved Genetic Programming Based Multiclass Classifier Using a New Crossover Operator
Game AI: artificial intelligence for 3D path finding
The role of artificial intelligence (AI) in games is gaining importance and often affects the suc... more The role of artificial intelligence (AI) in games is gaining importance and often affects the success or failure of a game. In this paper, we investigate the use of AI in game development. Research is done on how AI can be applied in games, and the advantages it brings along. As the fields of AI in game development are too wide to be covered, the focus of this project is placed on certain areas. Two programs are implemented through this project - (i) an intelligent camera system, and (ii) path-finding in a 3D application.
Auto-compensation of nonlinear influence of environmental parameters on the sensor characteristics using neural networks
Isa Transactions, 2005
Usually the environmental parameters influence the sensor characteristics in a nonlinear manner. ... more Usually the environmental parameters influence the sensor characteristics in a nonlinear manner. Therefore obtaining correct readout from a sensor under varying environmental conditions is a complex problem. In this paper we propose a neural network (NN)-based interface framework to automatically compensate for the nonlinear influence of the environmental temperature and the nonlinear-response characteristics of a capacitive pressure sensor (CPS) to provide correct readout. With extensive simulation studies we have shown that the NN-based inverse model of the CPS can estimate the applied pressure with a maximum error of +/- 1.0% for a wide temperature variation from 0 to 250 degrees C. A microcontroller unit-based implementation scheme is also proposed.
In this paper, we share some of our observations for designing and implementing a shopping mall d... more In this paper, we share some of our observations for designing and implementing a shopping mall design (called Hikegaya) with 3D technology. We constructed Hikegaya using Hypertext Preprocessor (PHP), MySQL, Adobe Photoshop 7, Discreet 3D Studio Max 6, and ParallelGraphics VrmlPad. The methodology for design is discussed. With the existing technologies and with our use for prototype implementation experience of Hikegaya, we conclude that 3D applications for E-commerce are feasible.
Cascaded Bidirectional Recurrent Neural Networks for Protein Secondary Structure Prediction
IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2007
Protein secondary structure (PSS) prediction is an important topic in bioinformatics. Our study o... more Protein secondary structure (PSS) prediction is an important topic in bioinformatics. Our study on a large set of non-homologous proteins shows that long-range interactions commonly exist and negatively affect PSS prediction. Besides, we also reveal strong correlations between secondary structure (SS) elements. In order to take into account the long-range interactions and SS-SS correlations, we propose a novel prediction system based on cascaded bidirectional recurrent neural network (BRNN). We compare the cascaded BRNN against another two BRNN architectures, namely the original BRNN architecture used for speech recognition as well as Pollastri's BRNN that was proposed for PSS prediction. Our cascaded BRNN achieves an overall three state accuracy Q3 of 74.38\%, and reaches a high Segment OVerlap (SOV) of 66.0455. It outperforms the original BRNN and Pollastri's BRNN in both Q3 and SOV. Specifically, it improves the SOV score by 4-6%.
IEEE Transactions on Fuzzy Systems, 2009
Learning of (context-free) grammar rules that are based on alignment between texts of a given col... more Learning of (context-free) grammar rules that are based on alignment between texts of a given collection of sentences has attracted the attention of many researchers. We define and study the alignment profile and formulate fuzzy similarity of alignment profiles for a given collection of sentences. Using the fuzzy-similarity-based profile alignment, we give a methodology to formulate stochastic context-free grammar (CFG) rules. We introduce profile-alignment-based dynamic sentence similarity threshold to formulate the rules of stochastic CFG. The proposed methodology is tested using Child Language Data Exchange System (CHILDES) dataset of sentences. The benefits of our approach are experimentally demonstrated. Since our approach does not make use of any domain knowledge, it is expected to be useful in wide variety of applications requiring model construction.
Performance evaluation of SVM based semi-supervised classification algorithm
To construct decision boundaries for two-class classification, SVM approach is attractive due to ... more To construct decision boundaries for two-class classification, SVM approach is attractive due to its efficiency. However, this approach is useful for 2-class classification and when the classes (labels) for the data are known. In practice, we have collection of labeled as well as unlabelled data, and it gives rise to semi-supervised classification problem. In this paper, we give a semi-supervised
We carry out a comprehensive study of long-range interactions on a large data set of non-homologo... more We carry out a comprehensive study of long-range interactions on a large data set of non-homologous proteins. Our study reveals that the long-range interactions between amino acids far apart are common in protein folding, and play an important role on the formation of secondary structure. Using residue-wise contact order(RWCO) to describe long-range interactions, we further evaluate the effect of long-range interactions on secondary structure prediction. We select six most popular prediction methods and collect their prediction results on the same set of proteins. All the six prediction methods show a significant negative correlation between prediction accuracy and RWCO.
Uploads
Papers by Narendra Chaudhari