Papers by Ajeesh Ramanujan
Lecture notes in networks and systems, 2024
xtraction of information from graph data is essential since most data in
the real world can be d... more xtraction of information from graph data is essential since most data in
the real world can be dynamic, large, and without fixed structure, unlike images.
Graph neural networks (GNNs) harness the power of graphs, efficiently examining
graph data and help in making inferences from complex data structures, making
them an invaluable tool in domains like social network analysis. GNNs have shown
promise in solving combinatorial optimization problems which involve extracting
a good candidate solution by scanning over the search space. The vastness of the
search space often renders the optimal solution search a difficult task. Many pre-
vious works demonstrate the use of GNNs in solving problems like graph coloring
and MaxCut problem [2, 11]. Following the work in the paper ”Combinatorial opti-
mization with physics-inspired graph neural networks” [14], the MaxCut problem is
posed as a Quadratic Unconstrained Binary(QUBO) problem, minimizing the loss
function based on the QUBO objective. In this paper we analyze the ability of differ-
ent GNN architectures to tackle the MaxCut problem via QUBO formulation setup.
Deep Learning Enabled Classification of Cognitive Impairment Stages Using MRI Images
2023 International Conference on Computational Intelligence and Sustainable Engineering Solutions (CISES)
A Survey on DDoS Prevention, Detection, and Traceback in Cloud
Transactions on Computational Science and Computational Intelligence, 2021
Distributed Denial of Service (DDoS) ranks among the top ten threats to the cloud computing envir... more Distributed Denial of Service (DDoS) ranks among the top ten threats to the cloud computing environment. DDoS mainly targets limited resources of cloud like bandwidth and CPU thereby denying access to legitimate clients. DDoS attacks are initiated by a vast network of remotely controlled nodes called zombies. New forms of DDoS are invented every day. Therefore, DDoS preventive measures do not fully guarantee its mitigation. Detecting an attack and defending it as early as possible is critical for reducing the attack impact. The real solution to mitigate any attack is tracing back the attacker and punishing him. However, a real attacker will masquerade his identity using a spoofed address to avoid being traced back. The routing mechanism used on the internet does not have any memory of its own making traceback further difficult. Many businesses are reluctant to enter the cloud due to these DDoS vulnerabilities of the cloud. DDoS will affect network performance and may disrupt configuration information available in the system. In the event of DDoS, businesses will have to suffer reputation damage, customer agitation, and legal repercussions. Unless cloud is made secure, we cannot benefit from its full potential. Research on DDoS attacks and defense is in its infancy. DDoS defense and traceback is still an open and challenging problem. This paper presents basic types of DDoS and focuses more on DDoS prevention, detection, and traceback techniques.
On Controlled P Systems
Eleventh Brainstorming Week on Membrane Computing, 2013, ISBN 9788494069192, págs. 137-151, 2013
Summary. We introduce and briefly investigate P systems with controlled computations. First, P sy... more Summary. We introduce and briefly investigate P systems with controlled computations. First, P systems with label restricted transitions are considered (in each step, all rules used have either the same label, or, possibly, the empty label, λ), then P systems with the computations controlled by languages (as in context-free controlled grammars). The relationships between the families of sets of numbers computed by the various classes of controlled P systems are investigated, also comparing them with length sets of languages in Chomsky and Lindenmayer hierarchies (characterizations of the length sets of ET0L and of recursively enumerable languages are obtained in this framework). A series of open problems and research topics are formulated. 1
On The Power Of Distributed Bottom-up Tree Automata
— Tree automata have been defined to accept trees. Different types of acceptance like bottom-up, ... more — Tree automata have been defined to accept trees. Different types of acceptance like bottom-up, top-down, tree walking have been considered in the literature. In this paper, we consider bottom-up tree automata and discuss the sequential distributed version of this model. Generally, this type of distribution is called cooperative distributed automata or the blackboard model. We define the traditional five modes of cooperation, viz. ∗-mode, t-mode, = k, ≥ k, ≤ k (k ≥ 1) modes on bottom-up tree automata. We discuss the accepting power of cooperative distributed tree automata under these modes of cooperation. We find that the ∗-mode does not increase the power, whereas the other modes increase the power. We discuss a few results comparing the acceptance power under different modes of cooperation.
We consider labeled spiking neural P systems, which are usual spiking neural P systems with a lab... more We consider labeled spiking neural P systems, which are usual spiking neural P systems with a label associated with every rule; the labels are symbols of a given alphabet or can be λ (empty). The rules used in a transition should have either the empty label or the same label from the chosen alphabet. In this way, a string is associated with each halting computation, called the control word of the computation. The set of all control words associated with computations in a given spiking neural P system form the control language of the system. We study the family of control languages of spiking neural P systems in comparison with the families of finite, regular, context-free, context-sensitive, and recursively enumerable languages. In the restricted case when in each step at least one rule with a non-empty label is used, every regular language is a control language, there are context-sensitive non-context-free languages of this type, but not all context-free languages are control langu...
International journal of advanced computer science, 2013
Tree automata have been defined to accept trees. Different types of acceptance like bottom-up, to... more Tree automata have been defined to accept trees. Different types of acceptance like bottom-up, top-down, tree walking have been considered in the literature. In this paper, we consider bottom-up tree automata and discuss the sequential distributed version of this model. Generally, this type of distribution is called cooperative distributed automata or the blackboard model. We define the traditional five modes of cooperation, viz. ∗-mode, t-mode, = k, ≥ k, ≤ k (k ≥ 1) modes on bottom-up tree automata. We discuss the accepting power of cooperative distributed tree automata under these modes of cooperation. We find that the ∗- mode does not increase the power, whereas the other modes increase the power. We discuss a few results comparing the acceptance power under different modes of cooperation.
Improving relation extraction beyond sentence boundaries using attention
Reation Extraction(RE) is the subprocess of Information Extraction(IE) which focuses on determini... more Reation Extraction(RE) is the subprocess of Information Extraction(IE) which focuses on determining and extracting the reation between two participating entities. Most of the past work focus on extracting relations within a sentence. Nowadays, research on relation extraction focuses on identifying and determining relationship between participating entities across sentences. This paper proposes a bi-directional GRU model with self attention mechanism for inter-sentential relation extraction. First, a bi-directional GRU with self attention mechanism is used to capture the information about the relation from intermediary terms between two entities. Then a bi-directional GRU is used to capture the information represented by entities, which plays a vital role in relation extraction. Finally, the proposed model combines both word embeddings and entity embeddings for extracting a relation. Experimental results show that the proposed Bi-directional GRU model can deliver state-of-the-art res...
Relation Extraction across sentences using Bi-directional Long Short Term Memory Networks
2020 International Conference on Emerging Trends in Information Technology and Engineering (ic-ETITE), 2020
Most of the past work on relation extraction(RE) has focused on identifying relationships between... more Most of the past work on relation extraction(RE) has focused on identifying relationships between entities within a sentence. Nowadays, most of the research in the field of RE has got interested in relation extraction between entity pairs across sentence boundaries. This paper proposes a Bi-directional LSTM model for for inter-sentential RE. Experimental results show that the proposed Bi-LSTM model can achieve better results on relation classification by capturing the information hidden in long-distance relation patterns.
A Survey on DDoS Prevention, Detection, and Traceback in Cloud
Distributed Denial of Service (DDoS) ranks among the top ten threats to the cloud computing envir... more Distributed Denial of Service (DDoS) ranks among the top ten threats to the cloud computing environment. DDoS mainly targets limited resources of cloud like bandwidth and CPU thereby denying access to legitimate clients. DDoS attacks are initiated by a vast network of remotely controlled nodes called zombies. New forms of DDoS are invented every day. Therefore, DDoS preventive measures do not fully guarantee its mitigation. Detecting an attack and defending it as early as possible is critical for reducing the attack impact. The real solution to mitigate any attack is tracing back the attacker and punishing him. However, a real attacker will masquerade his identity using a spoofed address to avoid being traced back. The routing mechanism used on the internet does not have any memory of its own making traceback further difficult. Many businesses are reluctant to enter the cloud due to these DDoS vulnerabilities of the cloud. DDoS will affect network performance and may disrupt configu...
Architecture of a Semantic WordCloud Visualization
Relation extraction has an important role in extracting structured information from unstructured ... more Relation extraction has an important role in extracting structured information from unstructured raw text. This task is a crucial ingredient in numerous information extraction systems seeking to mine structured facts from text. Nowadays, neural networks play an important role in the task of relation extraction. The traditional non deep learning models require feature engineering. Deep Learning models such as Convolutional Neural Networks and Long Short Term Memory networks require less feature engineering than non-deep learning models. Relation Extraction has the potential of employing deep learning models with the creation of huge datasets using distant supervision. This paper surveys the current trend in Relation Extraction using Deep Learning models.
Deep Learning Applications with Python
Advanced Deep Learning for Engineers and Scientists
The applications of deep learning extend to many aspects of daily life and are not confined to th... more The applications of deep learning extend to many aspects of daily life and are not confined to the domains of computer science alone. From face recognition to the smart grid domain, deep learning has proved itself to be an effective tool in producing state-of-the-art results. This chapter discusses how the usage of a high-level language like Python and its compatibility with deep learning frameworks and its collection of utility libraries facilitates practitioners in the development process. The chapter further dwells into details of the implementation of deep learning techniques applied to different applications, namely, facial recognition, fingerprint recognition, character recognition, smart grids, and renewable energy, by providing a brief history of how the technology has influenced the domain and details regarding the common datasets used as well as a code implementation in Python thoroughly covering the different steps.
Internet Censorship Based on Bayes Learning Model
Control languages accepted by labeled spiking neural P systems with rules on synapses
Theoretical Computer Science
On Controlled P Systems
Fundamenta Informaticae
ABSTRACT
International Journal of Advances in Engineering Sciences and Applied Mathematics
P systems with controlled computations have been introduced and investigated in the recent past, ... more P systems with controlled computations have been introduced and investigated in the recent past, by assigning labels to the rules in the regions of the P system and guiding the computations by control words. Here we consider string rewriting cell-like transition P system with label assigned rules working in acceptor mode and compare the obtained family of languages of control words over the rule labels with certain well-known language families. An application to chain code picture generation is also pointed out.
Malayalam text summarization: An extractive approach
2016 International Conference on Next Generation Intelligent Systems (ICNGIS), 2016
Ramanujan sums based image kernels for computer vision
2016 International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT), 2016
Uploads
Papers by Ajeesh Ramanujan
the real world can be dynamic, large, and without fixed structure, unlike images.
Graph neural networks (GNNs) harness the power of graphs, efficiently examining
graph data and help in making inferences from complex data structures, making
them an invaluable tool in domains like social network analysis. GNNs have shown
promise in solving combinatorial optimization problems which involve extracting
a good candidate solution by scanning over the search space. The vastness of the
search space often renders the optimal solution search a difficult task. Many pre-
vious works demonstrate the use of GNNs in solving problems like graph coloring
and MaxCut problem [2, 11]. Following the work in the paper ”Combinatorial opti-
mization with physics-inspired graph neural networks” [14], the MaxCut problem is
posed as a Quadratic Unconstrained Binary(QUBO) problem, minimizing the loss
function based on the QUBO objective. In this paper we analyze the ability of differ-
ent GNN architectures to tackle the MaxCut problem via QUBO formulation setup.