Faculty of Science, Technology & Environment


School of Computing, Information and Mathematical Sciences

Anurag Sharma
  • Anurag Sharma
    Job Info
    School of Computing, Information and Mathematical Sciences
    Position Title:
    Senior Lecturer
    Contact Info
    Phone #:
    +679 32 32816
  • Personal Info

    Postgraduate Course

    CS412 Artificial Intelligence
    CS415 Advanced Software Engineering
    Undergraduate Courses
    CS112 Data structures and Algorithms
    CS240 Software Engineering I
    CS241 Software Design & Implementation Data structures and Algorithms
    CS214 Design & Analysis of Algorithms
    CS218 Mobile Middleware
    CS324 Distributed Computing


    My research interest is manly in artificial intelligence and new optimization techniques in engineering specifically with the use of evolutionary algorithms. The brief description of my projects are given below:

    1. Enhancement of SGD algorithm for various machine learning models

    My current research focus is enhancement Stochastic Gradient Descent (SGD) algorithm for various machine learning models. SGD Algorithm, despite its simplicity, is considered an effective and default standard optimization algorithm for machine learning classification models such as neural networks and logistic regression. However, SGD’s gradient descent is biased towards the random selection (inconsistency) of a data instance. I proposed a Guided Stochastic Gradient Descent (GSGD) Algorithm to overcome this inconsistency. It has also been incorporated and tested with other popular variations of SGD, such as Adam, Adagrad and Momentum. Now we are focusing on parallelization of GSGD.

    2. Constraint guided search through evolutionary algorithm
    This is my PhD research topic which has been completed recently. I have been able to publish some quality papers as the outcome of this research. Many science and engineering applications require finding solutions to optimization problems by satisfying a set of constraints. These problems are typically ill-structured and intractable. They are formalized as constraint problems (CPs). Evolutionary algorithms (EAs) are known to be good solvers for optimization problems ubiquitous in various problem domains. EAs have also been used to solve CPs, however traditional EAs are ‘blind’ to constraints as they do not extract and exploit information from the constraints to better ‘inform’ search for solutions. I developed a variation of EA - Intelligent constraint handling evolutionary algorithm (ICHEA) that has been demonstrated to be a versatile constraints-guided EA for all forms of constrained problems on several benchmark problems. ICHEA works on domains with both quantitative and qualitative data and constraints. ICHEA modeling is designed to be independent of problem parameters and mostly focuses on maximally utilizing information from constraint in its search. It takes the divide and conquer approach for search through incrementally taking a sequence of constraints to solve CPs. It models constraints preferences and constraints strengths and exploits their relationships in search. The computational model developed allows for what-if analysis for exploring search spaces and for revising solution paths. Incrementality is a requisite feature as it models solving dynamic CPs where constraints arrive at run time or where constraints change over time. The challenge undertaken is to maximally utilize solutions already developed to a point in time to process any new, arriving sub-sequence of constraints rather than search for a solution anew each time a constraint changes or a new constraint arrives.

    3. Determine cluster boundaries using heuristic algorithms
    This has been the research of my Master’s Thesis. The title was Clustering for Data Mining: A Hybrid Particle Swarm Optimization – Self Organizing Map Approach where we have proposed a novel algorithm that uses Particle Swarm Optimization (PSO) algorithm to determine the cluster boundaries in the output of self-organizing map (SOM). SOM is a data mining tool that reveals structure in data sets through data visualization that is otherwise hard to detect from raw data alone. However, interpretation through visual inspection is prone to errors and can be very tedious. There are several techniques for the automatic detection of clusters of code vectors found by SOM, but they generally do not take into account the distribution of code vectors. The results of this novel automatic method compete very favorably to boundary detection through traditional algorithms namely k-means and hierarchical based approach which are normally used to interpret the output of SOM. Particle Swarm Optimization (PSO) is one of the newly developed algorithms being investigated internationally. This kind of algorithms solves those problems where problem formulation is either impossible or very time consuming to process.

    4. Some  recent projects for  MSc/PhD Students:

    i. Auto translation of iTaukei to English App  
    A student is doing a project on natural language processing (NLP) using AI tools. He is collecting audio data of different iTaukei dialects and using recurrent neural networks (RNN) for the translation of iTaukei to English and vice-versa. He would also be working on cloud analytics to have a generic auto data cleaning for text data sets. 

    ii. Image edge detection for dimension calculation for container loading
    Another student is working on image dimension calculations. Currently he is doing literature survey for the state-of-the-art tools for dimension calculation using mobile camera, however the current finding suggests that there are not many tools that are user-friendly as well as accurate enough to be acceptable. The work is in progress.

    iii. Improving Stochastic Gradient Descent (SGD) Algorithm
    During the teaching of the course CS412: Artificial Intelligence I figured out there is possibility of improving the well-known SGD algorithm using a guided approach based on consistency/inconsistency of the dataset. A paper has been accepted in Applied Soft Computing Journal.

    iv. Bilevel Optimization
    Recently, there has been an increased interest from evolutionary computation community to model bilevel problems due to its applicability in the real world applications for decision-making problems. In this work, we are working on a non-nested complete evolutionary approach to solve the benchmark problems. We are also proposing new variants to the commonly used convergence approaches, i.e., optimistic and pessimistic. The experimental results demonstrate the algorithm converges differently to known optimum solutions with the optimistic variants. A paper with very promising results is under submission.

    v. Skill based group allocation of students for project-based learning courses
    I have started working on this project with my PhD student. We have already submitted a paper to ICONIP 2018. Conventionally, facilitators assign the students to the group randomly which results in biased groups where all the necessary skills to complete the project lacks in some of the groups. Most computational tools solves the Group Assignment Problem (GAP) by assigning students to relevant groups based on some general criterion such as maximizing the diversity of the group members. However, there is a need for a system which allows to take skill preference as a parameter in a limited or unevenly distributed skill set. The system needs to have more or less same strength with presence of all the skills required to complete the project successfully. We have proposed a nature-inspired heuristic method that finds evenly balanced groups by minimizing inter group difference in terms of allocated skill set. Promising results have been obtained.

    vi. Auto digit recognition improvement through a novel preprocessing technique
    Norman Bentley was working on this. He has already achieved some preliminary results which are promising. The preprocessing technique is the hybridization of Artificial Neural Network with Simulated Annealing algorithm.

    vii. Usefulness of transfer learning in English accent recognition
    Krishan Kumar is currently evaluating the advantages to transfer learning in recognizing 14 different English accents from a benchmark dataset. Currently transfer learning is only confined to deep neural networks.

    viii. Feasibility of Smart E-commerce in Fiji for online shopping 

    An IS student has started working on this topic. His finding would determine the feasibility of e-commerce in Fiji supported by AI tools to have optimum delivery methods to customers, determination of most profitable pricing for shop owners, and knowledge discovery on the behavior of products consumption. 

    Publications in USP Electronic Research Repository


    I have 2 years of experience as a professional IT programmer. I worked for Colonial Fiji to develop and maintain many sofware systems such as Health Insurance System, Credit Card System and Cashier System.


    For citation information, please refer to my Google Scholar profile.
    Book Chapters

    1. A. Sharma and G. Onwubolu, Hybrid Particle Swarm Optimization and GMDH System, In: (ed. Onwubolu, G. C.) Hybrid Self-Organizing Modeling Systems, Springer-Verlag, Germany, 2009.
    2. G. C. Onwubolu and A. Sharma, Particle Swarm Optimization for the assignment of facilities to locations. New Optimization Techniques in Engineering, Springer-Verlag, 2004.

    Conference Proceedings

    1. A. Sharma, “Analysis of Evolutionary Operators for ICHEA in Solving Constraint Optimization Problems,” presented at the IEEE CEC 2015, Sendai, Japan, 2015, pp. 46–53 (ERA – A)

    2. A. Sharma and D. Sharma, ICHEA – A Constraint Guided Search for Improving Evolutionary Algorithms, in Neural Information Processing, vol. 7663, Huang, T., Zeng, Z., Li, C., Leung, C.S. (eds.) ICONIP 2012, Part I. LNCS, Doha, Qatar, 2012, pp. 269–279. (ERA – A)

    3. A. Sharma and D. Sharma, Real-Valued Constraint Optimization with ICHEA, in Neural Information Processing, vol. 7665, T. Huang, Z. Zeng, C. Li, and C. Leung, Eds. Springer Berlin / Heidelberg, 2012, pp. 406–416. (ERA – A)

    4. A. Sharma and D. Sharma, Solving Dynamic Constraint Optimization Problems Using ICHEA, in Neural Information Processing, vol. 7665, T. Huang, Z. Zeng, C. Li, and C. Leung, Eds. Springer Berlin / Heidelberg, 2012, pp. 434–444. (ERA – A)

    5. A. Sharma and D. Sharma, An Incremental Approach to Solving Dynamic Constraint Satisfaction Problems, in Neural Information Processing, vol. 7665, T. Huang, Z. Zeng, C. Li, and C. Leung, Eds. Springer Berlin / Heidelberg, 2012, pp. 445–455. (ERA A)

    6. A. Sharma and D. Sharma, Clonal Selection Algorithm for Classification, in Artificial Immune Systems, vol. 6825, P. Liò, G. Nicosia, and T. Stibor, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2011, pp. 361–370.

    7. A. Sharma, “Analysis of Evolutionary Operators for ICHEA in Solving Constraint Optimization Problems,” in IEEE Congress on Evolutionary Computation (CEC), Sendai, Japan, 2015, pp. 46–53. (ERA A)

    8. S. Hussain, R. Chandra and A. Sharma, “Multi-Step-Ahead Chaotic Time Series Prediction using Coevolutionary Recurrent Neural Networks,” in IEEE Congress on Evolutionary Computation (CEC), Vancouver, Canada, 2016 (ERA A)

    9. G. Wong, A. Sharma and R. Chandra, “Information Collection Strategies in Memetic Cooperative Neuroevolution for Time Series Prediction”, in IJCNN, Rio, Brazil, 2018 (ERA A)

    10. A. K. Sharma, V. Prasad, R. Kumar and A. Sharma, Analysis on the Occurrence of Tropical cyclone in the South Pacific Region using Recurrent Neural Network with LSTM, ICONIP, 2018 (Accepted) (ERA A)

    11. R. Nand and A. Sharma, Skill Based Group Allocation of Students for Project-Based Learning Courses using Genetic Algorithm, part I: Weighted Penalty Model, IEEE TALE, Wollongong, Australia, 2018 (Accepted) (USP A)


    12. A. Sharma and C. W. Omlin, Performance comparison of Particle Swarm Optimization with traditional clustering algorithms used in Self Organizing Map, International Journal of Computational Intelligence, vol. 5, No. 1.1, pp. 1-12, 2008.

    13. A. Sharma, “Guided Stochastic Gradient Descent Algorithm for inconsistent datasets,” Applied Soft Computing, vol. 73, pp. 1068–1080, Dec. 2018 (USP A)

    14. A. Sharma, Optimistic Variants of Single-Objective Bilevel Optimization for Evolutionary Algorithms (under submission).


    1. Most cited and downloaded paper in Artificial Immune Systems – 2011: A. Sharma and D. Sharma, Clonal Selection Algorithm for Classification, in Artificial Immune Systems, vol. 6825, P. Liò, G. Nicosia, and T. Stibor, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2011, pp. 361–370.
    2. International Postgraduate Research Scholarship (IPRS) to do PhD in Australia (2010).
    3. Gold Medal for most outstanding Master’s thesis at the University of the South Pacific, Fiji (2008).
    4. Received several awards for solving complex mathematical problems at the University of the South Pacific, Fiji (2000-2002).


Disclaimer & Copyright l Contact Us l 
© Copyright 1968 - 2019. All Rights Reserved.
Page updated: Wednesday, May 22, 2019
Faculty of Science, Technology & Environment
The University of the South Pacific
Private Bag, Laucala Campus,
Suva, Fiji.
Tel: (+679) 323 1000
Fax: (+679) 323 1506