Papers by Sriram. G. Sanjeevi
International Journal of Reasoning-based Intelligent Systems, 2015
In rough set theory, attribute reduction is an important application. Many approaches to attribut... more In rough set theory, attribute reduction is an important application. Many approaches to attribute reduction were developed using rough set theory. These approaches used various greedy heuristics to find a reduct approximation. In this paper, we propose forward tentative selection with backward propagation of selection decision (FTSBPSD) algorithm to find a reduct. The proposed algorithm is based on the principle of indiscernibility of rough set theory. It finds one of the prime implicants of the discernibility function as the reduct. The proposed algorithm works for various types of reducts, defined in the rough set theory. In this work, we have analysed the performance of distribution reduct, maximum distribution reduct, positive region reduct and possible reduct. The proposed algorithm was tested on various datasets found in University of California, machine learning repository. It has given good results for classification accuracy during tests performed on the datasets. Experimental results obtained by FTSBPSD algorithm have been found to give better classification accuracy when tested using C4.5 classifier in comparison to the results obtained by the Q-MDRA algorithm described in the literature.
2021 International Conference on Communication, Control and Information Sciences (ICCISc), 2021
Recent work in abstractive text summarization using pre-trained transformers has achieved great r... more Recent work in abstractive text summarization using pre-trained transformers has achieved great results. Much of the work has been done on the model architectures and designing pre-training objectives. Models used for NLP tasks have grown bigger and increasing the model’s size doesn’t always improve the performance of the task. In this work we improve the results of an existing model by incorporating a new fine-tuning strategy which closely resembles how a human would summarize by a piece of text i.e. by focusing on important fragments of the text. We used PEGASUS model to work on our fine-tuning strategy. We fine-tune the model with varying ratios of important sentences in a phased manner and achieve improved ROUGE scores with an additional few thousand steps of fine-tuning. We fine-tune the model on AESLC and CNN/DailyMail dataset and achieve better ROUGE scores on both the datasets. With our proposed fine-tuning method, we get state-of-the-art results on the AESLC dataset. We also achieve better performance when fine-tuning with a limited number of examples with our fine-tuning method on AESLC dataset.
2021 International Conference on Communication, Control and Information Sciences (ICCISc), 2021
Image captioning models are in abundance, with various techniques and methods in place for them. ... more Image captioning models are in abundance, with various techniques and methods in place for them. One such primitive model is "Show, Attend and Tell". This model showcases how well attention mechanisms can be used to focus on necessary parts of an image to predict the next word in the caption. While this model uses soft attention mechanism, the recent development of transformer introduced the concept of multi-head attention mechanism, which was the inspiration for using it instead of soft attention. Multi-head attention was primarily introduced to work on text, but we experimented it on images and text here. We show different components of the captioning model such as the encoder (ResNet), decoder (LSTM with Bert Embeddings) and the use of multi-head attention for image and text instead of text only (as previously used in transformer). We work on the benchmark dataset of MSCOCO. We prepared a baseline and achieved better BLEU score results with our proposed model.
Recommendation systems are the systems that recommend items to the users based on their tastes an... more Recommendation systems are the systems that recommend items to the users based on their tastes and interest. With the expansion of data on the web and a variety of products available in the E-Commerce sector, the recommender systems come to the aid of the users. The cold start problem is one of the major drawbacks for the recommender system and can prove costly for its success if proper steps are not taken. We solve this by proposing variations in adaptive clustering methods that will be used in the pre-processing stage to generate data only for lowly ranked items so that they can compete with the highly ranked items. We also develop a predictor based on incremental learning algorithm based on LEARN++ that meets the needs of the adaptively clustered data. Keywords—Adaptive Clustering, Cold Start, Incremental learning, Recommendation System, Slope One predictor
Proceedings of International Conference on Computational Intelligence and Data Engineering, 2020
High utility itemsets are having utility more than user-specified minimum utility. These itemsets... more High utility itemsets are having utility more than user-specified minimum utility. These itemsets provide high profit but do not exhibit correlation between them. High utility itemsets mining generate huge number of itemsets considering only single interesting criteria. Existing algorithm mines correlated high utility itemsets mining extracts itemsets that provide high utility with correlation between them. The limitation of this algorithm is that it does not consider the rarity of itemsets. To overcome this limitation, this proposed algorithm mines rare correlated high utility itemsets. Firstly, it mines correlated high utility itemsets. Secondly, it determines whether the itemsets support is no greater than minsup specified by the user. It can be shown by experimental results that the proposed algorithm reduces considerably runtime and number of candidate itemsets.
8th ACM IKDD CODS and 26th COMAD, 2021
Stock portfolio allocation is one of the most challenging and interesting problems of modern fina... more Stock portfolio allocation is one of the most challenging and interesting problems of modern finance. Recently, deep reinforcement learning applications have shown promising results in automating portfolio allocation. However, most current approaches use a single agent learning model which could inadequately capture the complex dynamics arising from the interactions of many traders in today’s stock market. In this paper, we explore the applicability of multi-agent deep reinforcement learning to this problem by implementing single-agent, 2-agent, 3-agent, and 4-agent deep deterministic policy gradients (DDPG) algorithms in a competitive setting. Upon analyzing the results obtained using standardized metrics, we observe that there is a significant improvement in the performance of our learning models with the introduction of multiple agents.
Engineering Science and Technology, an International Journal, 2021
Abstract The stock market currently remains one of the most difficult systems to model in finance... more Abstract The stock market currently remains one of the most difficult systems to model in finance. Hence, it is a challenge to solve stock portfolio allocation wherein an optimal investment strategy must be found for a curated collection of stocks that effectively maximizes return while minimizing the risk involved. Deep reinforcement learning approaches have shown promising results when used to automate portfolio allocation, by training an intelligent agent on historical stock prices. However, modern investors are actively engaging with digital platforms such as social media and online news websites to understand and better analyze portfolios. The overall attitude thus formed by investors toward a particular stock or financial market is known as market sentiment. Existing approaches do not incorporate market sentiment which has been empirically shown to influence investor decisions. In our paper, we propose a novel deep reinforcement learning approach to effectively train an intelligent automated trader, that not only uses the historical stock price data but also perceives market sentiment for a stock portfolio consisting of the Dow Jones companies. We demonstrate that our approach is more robust in comparison to existing baselines across standardized metrics such as the Sharpe ratio and annualized investment return.
International Journal of Recent Technology and Engineering (IJRTE), 2019
High Utility Item sets mining has attracted many researchers in recent years. But HUI mining meth... more High Utility Item sets mining has attracted many researchers in recent years. But HUI mining methods involves a exponential mining space and returns a very large number of high-utility itemsets. . Temporal periodicity of itemset is considered recently as an important interesting criteria for mining high-utility itemsets in many applications. Periodic High Utility item sets mining methods has a limitation that it does not consider frequency and not suitable for large databases. To address this problem, we have proposed two efficient algorithms named FPHUI( mining periodic frequent HUIs), MFPHM(efficient mining periodic frequent HUIs) for mining periodic frequent high-utility itemsets. The first algorithm FPHUI miner generates all periodic frequent itemsets. Mining periodic frequent high-utility itemsets leads to more computational cost in very large databases. We further developed another algorithm called MFPHM to overcome this limitation. The performance of the frequent FPHUI miner ...
Soft Computing, 2016
Discretization and attribute reduction are two preprocessing steps for most of the induction algo... more Discretization and attribute reduction are two preprocessing steps for most of the induction algorithms. Discretization before attribute reduction will result in high computation cost as many irrelevant and redundant attributes need to be discretized. Attribute reduction before discretization may result in over-fitting of the data leading to low performance of the induction algorithm. In this paper, we have proposed a hybrid algorithm using artificial bee colony (ABC) algorithm and extended forward tentative selection with backward propagation of selection decision (EFTS-BPSD) algorithm for attribute reduction on real-valued data in rough set theory (RST). Based on the principle of indiscernibility, the hybrid ABC-EFTSBPSD algorithm performs discretization and attribute reduction together. The hybrid ABC-EFTSBPSD algorithm takes as input the decision system consisting of real-valued attributes and determines a near optimal set of irreducible cuts. Here, optimality of the set of irreducible cuts is defined in terms of the cardinality of the set of irreducible cuts. Reduct is obtained from the determined approximate optimal set of irreducible cuts by extracting the attributes corresponding to the cuts in the obtained set of irreducible cuts. The proposed hybrid algorithm is tested on various data sets from University of California Machine Learning Repository. Experimental results obtained by the proposed hybrid algorithm are compared with those obtained by the Q-MDRA, ACO-RST and Communicated by V. Loia.
Proceedings of the Third International Symposium on Women in Computing and Informatics - WCI '15, 2015
Attribute reduction techniques based on Pawlak rough set theory work only on data sets with discr... more Attribute reduction techniques based on Pawlak rough set theory work only on data sets with discrete attributes. In real-world applications, the domain of a few or all attributes of the data set may be continuous. These continuous attributes need to be discretized as a pre-processing step to attribute reduction. In this paper, we have proposed an algorithm to the problem of attribute reduction on continuous data in rough set theory. The proposed algorithm does not need any extra information or expert domain knowledge apart from the continuous data set. The proposed algorithm is based on the concepts of rough set theory. These include principle of indiscernibility, basic cuts and discernibility matrix. It adapts the search techniques provided by the ant colony optimization meta-heuristic. As ant colony optimization is a graph based meta-heuristic algorithm, we have introduced a fully connected graph whose nodes are the basic cuts. We have evaluated the proposed algorithm on various data sets found in University of California, machine learning repository. For each data set, a reduced data set is obtained by retaining the attributes in the reduct determined by the proposed algorithm and removing the attributes not in the reduct. The obtained reduced data set is found to give better classification accuracies when tested using i) C4.5 classifier and ii) Naive Bayes classifier in comparison with those obtained on the data set before attribute reduction.
Proceedings of the International Conference and Workshop on Emerging Trends in Technology - ICWET '10, 2010
... Sanjeevi, SG, and Bhattacharyya, P., A fault tolerant connectionist model for predicate logic... more ... Sanjeevi, SG, and Bhattacharyya, P., A fault tolerant connectionist model for predicate logic reasoning, variable binding: using coarse-coded distributed representations. WSEAS Transactions on Systems, 4, 4, (Apr. 2005), 331-336. [6] Shastri, L. Advances in SHRUTI: a neurally ...
Proceedings of the International Conference and Workshop on Emerging Trends in Technology, 2010
In this paper, we describe a fault-tolerant Neuro-Fuzzy inference system for performing fuzzy rea... more In this paper, we describe a fault-tolerant Neuro-Fuzzy inference system for performing fuzzy reasoning using coarse-coded distributed representations. The system implements the fuzzy membership functions in a novel way using coarse-coded distributed representations for the inputs and outputs of neural networks. Distributed representations are known to give advantages of fault tolerance, generalization and graceful degradation of performance under noise conditions. Performance of the Neuro-Fuzzy inference system with regard to its ability to exhibit fault tolerance under noise conditions is studied. The system offered very good results of fault tolerance under noise conditions. It has also exhibited good generalization ability on unseen test inputs.
In this work, we propose and present a Hybrid particle swarm optimization-Simulated annealing alg... more In this work, we propose and present a Hybrid particle swarm optimization-Simulated annealing algorithm and compare it with a Genetic algorithm for training respectively neural networks of identical architectures. These neural networks were then tested on a classification task. In particle swarm optimization, behavior of a particle is influenced by the experiential knowledge of the particle as well as socially exchanged information. Particle swarm optimization follows a parallel search strategy. In simulated annealing uphill moves are made in the search space in a stochastic fashion in addition to the downhill moves. Simulated annealing therefore has better scope of escaping local minima and reach a global minimum in the search space. Thus simulated annealing gives a selective randomness to the search. Genetic algorithm performs parallel and randomized search. The goal of training the neural network is to minimize the sum of the squares of the error between the target and observed output values for all the training samples and to deliver good test performance on the test inputs. We compared the performance of the neural networks of identical architectures trained by the Hybrid particle swarm optimization-simulated annealing and Genetic algorithm respectively on a classification task and noted the results obtained. Neural network trained by Hybrid particle swarm optimization-simulated annealing has given better results compared to the neural network trained by the Genetic algorithm in the tests conducted by us.
Smart Innovation, Systems and Technologies, 2011
Web Usage Mining is a broad area of Web Mining which is associated with the Patterns extraction f... more Web Usage Mining is a broad area of Web Mining which is associated with the Patterns extraction from logging information produced by web server. Web log mining is substantially the important part of Web Usage Mining (WUM) algorithm which involves transformation and interpretation of the logging information to predict the patterns as per different learning styles. Ultimately these patterns are
2010 IEEE International Conference on Computational Intelligence and Computing Research, 2010
ABSTRACT Traditionally e-learning systems are emphasized on the online content generation and mos... more ABSTRACT Traditionally e-learning systems are emphasized on the online content generation and most of them fail in considering the requirements and learning styles of end user, while representing it. Therefore, appears the need for adaptation to the user's learning behavior. Adaptive e-learning refers to an educational system that understands the learning content and the user interface according to pedagogical aspects. End users have unique ways of learning which may directly and indirectly affect on the learning process and its outcome. In order to implement effective and efficient e-learning, the system should be capable not only in adapting the content of course to the individual characteristics of students but also concentrate on the adaptive user interface according to students' requirements. In this paper, at initial stage we are presenting an approach to recognize the learning styles of individual student according to the actions or navigations that he or she has performed on an e-learning application. This recognition technique is based on Machine Learning algorithm called Artificial Neural Networks and Web Usage Mining.
International Journal of Computer Science, Engineering and Applications, 2011
In this work, we propose a Hybrid particle swarm optimization-Simulated annealing algorithm and p... more In this work, we propose a Hybrid particle swarm optimization-Simulated annealing algorithm and present a comparison with i) Simulated annealing algorithm and ii) Back propagation algorithm for training neural networks. These neural networks were then tested on a classification task. In particle swarm optimization behaviour of a particle is influenced by the experiential knowledge of the particle as well as socially exchanged information. Particle swarm optimization follows a parallel search strategy. In simulated annealing uphill moves are made in the search space in a stochastic fashion in addition to the downhill moves. Simulated annealing therefore has better scope of escaping local minima and reach a global minimum in the search space. Thus simulated annealing gives a selective randomness to the search. Back propagation algorithm uses gradient descent approach search for minimizing the error. Our goal of global minima in the task being done here is to come to lowest energy state, where energy state is being modelled as the sum of the squares of the error between the target and observed output values for all the training samples. We compared the performance of the neural networks of identical architectures trained by the i) Hybrid particle swarm optimization-simulated annealing, ii) Simulated annealing and iii) Back propagation algorithms respectively on a classification task and noted the results obtained. Neural network trained by Hybrid particle swarm optimization-simulated annealing has given better results compared to the neural networks trained by the Simulated annealing and Back propagation algorithms in the tests conducted by us.
2006 International Conference on Machine Learning and Cybernetics, 2006
... [2] Lokendra Shastri, and C. Wendelken, Multiple instantiation and rule mediation in SHRUTI... more ... [2] Lokendra Shastri, and C. Wendelken, Multiple instantiation and rule mediation in SHRUTI, Connection Science ... G. Sanjeevi, and P. Bhattacharyya, Connectionist Reasoning System using Coarse-coded Distributed Representations, Proceeding of ICSCI2005 Conference ...
2014 IEEE International Advance Computing Conference (IACC), 2014
ABSTRACT
Uploads
Papers by Sriram. G. Sanjeevi