Atul Sajjanhar’s research while affiliated with Deakin University and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (89)


A Unified Solution to Diverse Heterogeneities in One-shot Federated Learning
  • Preprint

October 2024

Jun Bai

·

Yiliao Song

·

·

[...]

·

Yan Li

One-shot federated learning (FL) limits the communication between the server and clients to a single round, which largely decreases the privacy leakage risks in traditional FLs requiring multiple communications. However, we find existing one-shot FL frameworks are vulnerable to distributional heterogeneity due to their insufficient focus on data heterogeneity while concentrating predominantly on model heterogeneity. Filling this gap, we propose a unified, data-free, one-shot federated learning framework (FedHydra) that can effectively address both model and data heterogeneity. Rather than applying existing value-only learning mechanisms, a structure-value learning mechanism is proposed in FedHydra. Specifically, a new stratified learning structure is proposed to cover data heterogeneity, and the value of each item during computation reflects model heterogeneity. By this design, the data and model heterogeneity issues are simultaneously monitored from different aspects during learning. Consequently, FedHydra can effectively mitigate both issues by minimizing their inherent conflicts. We compared FedHydra with three SOTA baselines on four benchmark datasets. Experimental results show that our method outperforms the previous one-shot FL methods in both homogeneous and heterogeneous settings.



Fig. 3: FL with local GAN augmentation for ITD. As the first step in FedAT, the GAN training is employed by the G model and the C model. Later, the C model is used for multiclass classification. Finally, the gradients from the C model and the auxiliary information (the class labels) are shared with the server.
Fig. 4: Network architectures of the SNN-MLP vs the classical MLP. (a) depicts the proposed SNN-MLP model, where SELU activation and AlphaDropout are adopted. (b) illustrates the adoption of the ReLU activation and the standard Dropout in the classical MLP model.
Fig. 5: Non-IID data distribution for the CERT datasets.
Fig. 6: Round-by-round testing performance and training loss.
Fig. 7: Effect of the local epochs on FedAT.
FedAT: Federated Adversarial Training for Distributed Insider Threat Detection
  • Preprint
  • File available

September 2024

·

21 Reads

Insider threats usually occur from within the workplace, where the attacker is an entity closely associated with the organization. The sequence of actions the entities take on the resources to which they have access rights allows us to identify the insiders. Insider Threat Detection (ITD) using Machine Learning (ML)-based approaches gained attention in the last few years. However, most techniques employed centralized ML methods to perform such an ITD. Organizations operating from multiple locations cannot contribute to the centralized models as the data is generated from various locations. In particular, the user behavior data, which is the primary source of ITD, cannot be shared among the locations due to privacy concerns. Additionally, the data distributed across various locations result in extreme class imbalance due to the rarity of attacks. Federated Learning (FL), a distributed data modeling paradigm, gained much interest recently. However, FL-enabled ITD is not yet explored, and it still needs research to study the significant issues of its implementation in practical settings. As such, our work investigates an FL-enabled multiclass ITD paradigm that considers non-Independent and Identically Distributed (non-IID) data distribution to detect insider threats from different locations (clients) of an organization. Specifically, we propose a Federated Adversarial Training (FedAT) approach using a generative model to alleviate the extreme data skewness arising from the non-IID data distribution among the clients. Besides, we propose to utilize a Self-normalized Neural Network-based Multi-Layer Perceptron (SNN-MLP) model to improve ITD. We perform comprehensive experiments and compare the results with the benchmarks to manifest the enhanced performance of the proposed FedATdriven ITD scheme.

Download




Figure 1. The study used a 4-stage process to gather and analyse qualitative and quantitative data to answer the research questions.
How teachers mitigate students' disengagement.
How teachers measure students' disengagement.
Teachers’ Perspectives on Using Augmented-Reality-Enhanced Analytics as a Measure of Student Disengagement

December 2023

·

60 Reads

Multimodal Technologies and Interaction

There are various ways that teachers manage student disengagement levels during their class lessons, and managing disengagement can be both stressful and challenging, especially since each student is unique. Methods and techniques utilised are specific to teachers’ own experience level, subject knowledge, and teaching styles. We report on the techniques and methods teachers utilise to identify, mitigate, and measure student disengagement during class lessons; the paper presents the results of a mixed-methods, multisession study design comprising gathered qualitative and quantitative data to enable a greater understanding. Eight educators who were full-time educators with varying years of experience from three different schools, who taught or had taught English, maths, and science subjects at the primary school level, participated in this study. The study also observed that teachers used three AR applications and collected valuable feedback on their perspectives by using analytics generated by AR applications to help manage student disengagement. A postsession survey tool was used to gather the perceived importance and ranking of the techniques and methods discussed by the teachers during the previous sessions. The results showed that the majority of teachers deemed spending “Time on Tasks” and giving “Feedback/Reflections” most suited for measuring disengagement, and encouraging “Movement” and use of “Technology” emerged as the most favoured for mitigating disengagement. For utilising AR enhanced analytics in mitigating and measuring student disengagement, the data suggested a difference in perspectives based on teachers’ teaching levels, especially concerning conversations and the use of technology devices. The study did not find conclusive evidence of differences based on teachers’ teaching subjects and there was a notable distinction in building positive relationships among English teachers. This leads to the suggestion that subject-specific pedagogy might influence the perceived effectiveness of using AR-generated analytics in mitigating and measuring student disengagement.


An Elastic Scalable Grouping for Stateful Operators in Stream Computing Systems

November 2023

·

10 Reads

Lecture Notes in Computer Science

In distributed stream computing systems, dynamic data skew and cluster heterogeneity can lead to major load imbalance among multiple instances of stateful operators. Existing stream grouping schemes mainly focus on data load balancing for stateful operators, but they are not considered to be sufficiently elastic scalable, which directly affects the latency and throughput. We propose an elastic scalable grouping (called Es-Stream) for stateful operators. This paper discusses the following aspects: (1) Investigating the dynamic grouping of real-time data stream, proposing a general data stream graph model and a data stream grouping model, as well as formalizing the problem of load balancing optimization and data stream grouping. (2) Utilizing key splitting to solve the bottleneck problem caused by high-frequency keys in the data streams, and lightweight weight adjustment strategy to dynamically change the data tuple allocation probability of the instance according to the network cost, data stream rate and processing rate. (3) Implementing Es-Stream in Apache Storm platform and evaluating the system using metrics such as latency, throughput and load imbalance. Experimental results showed that Es-Stream reduces latency by up to 72%, increases throughput by up to 44% and reduces load imbalance by up to 75%, compared with existing state-of-the-art grouping schemes.


Client Selection Based on Diversity Scaling for Federated Learning on Non-IID Data

July 2023

·

8 Reads

Lecture Notes of the Institute for Computer Sciences

In a wireless Federated Learning (FL) system, clients train their local models over local datasets on IoT devices. The derived local models are uploaded to the FL server which generates a global model, then broadcasts the model back to the clients for further training. Due to the heterogeneous feature of clients, client selection plays an important role in determining the overall training time. Traditionally, maximum number of clients are selected if they can derive and upload their local models before the deadline in each global iteration. However, selecting more clients not only increases the energy consumption of the clients, but also might not be necessary as having fewer clients in early global iterations and more clients in later iterations have been proved better for model accuracy. To address the issue, this paper proposes a client selection scheme which dynamically adjusts and optimizes the trade-off between maximizing the number of selected clients and minimizing the total communication cost between the clients and the server. By comparing the data diversity of clients, this scheme can select the most suitable clients for global convergence. A Diversity Scaling Node Selection framework (FedDS) is implemented to dynamically change the selecting weights of each node based on the degree of non-i.i.d data diversity. Results has shown that the proposed FedDS can speed up the FL convergence rate compared to FedAvg with random node selection.KeywordsFederated LearningDiversity ScalingConvergenceclient Selection


A two-tier coordinated load balancing strategy over skewed data streams

June 2023

·

79 Reads

·

3 Citations

The Journal of Supercomputing

Load imbalance severely affects cluster performance, and the polarization of resources due to load skewing leads to further worsening of system throughput and latency problems. The proliferation of tasks to be processed in the big data era leads to more severe load skewing. How to cope with the surge of skewed data stream in the context of big data is a new challenge now. In this paper, we propose a coordinated load balancing strategy on skewed data streams (referred to as St-Stream), which is a two-tier hierarchical system for handling data streams. The proposed strategy is characterized by performing a migration pairing strategy for resources at the task allocation stage by cutting and moving out the tasks of high-load nodes in a hierarchical manner, and the moved-out operators are placed in the routing table, and the routing table operators are moved out to these nodes sequentially according to the tasks required by low-load nodes. We further design a two-tier coordination scheme for the resource allocation problem, which can adjust the skewed load from within the nodes and then dynamically restore the balance between the nodes. We implemented St-Stream on Apache Storm, which achieves a 21% coordination in processing CPU utilization, a 17.6% reduction in latency, and a 0.3 improvement in load balance recovery compared to the baseline design. Our experimental results demonstrate that the proposed load balancing strategy better balances the cluster load and improves the performance of the stream processing system.


Citations (49)


... Autoencoders excel at capturing complex patterns but require significant computational resources. SVMs deliver high precision in smaller datasets but are less practical for large-scale applications [37]. ...

Reference:

Integrated Strategies for Database Protection: Leveraging Anomaly Detection and Predictive Modelling to Prevent Data Breaches
Hybrid deep learning model using SPCAGAN augmentation for insider threat analysis
  • Citing Article
  • February 2024

Expert Systems with Applications

... For instance, Ref. [21] presented an NFT-based model aimed at enhancing the copyright traceability of off-chain data, contributing to the sustainability of the NFT community. In a different application, Ref. [22] proposed a framework, KD-NFT, that integrates NFT security features with knowledge distillations to address security concerns. This model extends NFT security into machine learning, leveraging blockchain features to recover the training procedure. ...

A Robust NFT Assisted Knowledge Distillation Framework for Edge Computing
  • Citing Chapter
  • June 2023

Lecture Notes of the Institute for Computer Sciences

... A distinção entre RA e RV é importante na escolha de uma aplicação para determinado uso. Por exemplo, na educação, a RA pode ser usada para enriquecer o aprendizado com informações adicionais em tempo real [Sahin et al. 2018, Singh et al. 2023, como no estudo de anatomia, onde modelos 3D podem ser sobrepostos ao corpo humano para uma melhor compreensão. Por outro lado, a RV pode ser mais eficaz em simulações onde é necessário criar um ambiente completamente controlado e imersivo, como em treinamentos militares ou médicos [Elliman et al. 2016, Basdogan et al. 2007]. ...

Forest Classroom: A Case Study of Educational Augmented Reality Design to Facilitate Classroom Engagement

Multimodal Technologies and Interaction

... Nevertheless, there still needs to be more research in academic publications about the effectiveness of deep learning in learning analytics, especially when anticipatorily resolving student performance problems. Several ANN models, including long short-term memory (LSTM) and recurrent neural networks (RNN), are used to improve student outcomes proactively [25]. These models analyze learners' performance across daily, weekly, or monthly intervals, using the course schedule as a data series. ...

Predicting Student Performance Using Clickstream Data and Machine Learning

Education Sciences

... Singh et al. [74] also explored how to capture, analyze, and present learning data that derived from augmented reality applications. In their work, they presented a system architecture for mobile devices and applications which consisted of various parts, such as data collection, aggregation, and storage, process metrics, analytics presentation, learning outcomes, and affective factors analytics. ...

An Architecture for Capturing and Presenting Learning Outcomes using Augmented Reality Enhanced Analytics
  • Citing Conference Paper
  • October 2022

... Singh et al. [90] explored how augmented reality-enhanced analytics can be used to measure young learners' engagement. Their proposed system uses various learning analytics components and integrates them into an augmented reality application. ...

Augmented Reality Enhanced Analytics to Measure and Mitigate Disengagement in Teaching Young Children
  • Citing Conference Paper
  • October 2022

... Federated learning (FL) has become a machine learning paradigm [4], which can train models on various decentralized devices or servers, and it is a solution for constructing a jointtrained model while preserving data privacy and participants can keep data locally without exchanging [5]. Federated learning was proposed for the first time by McMahan et al. [6], the learning task settled by loose jointing of participating devices (called clients) coordinated by a central server. ...

FedEWA: Federated Learning with Elastic Weighted Averaging
  • Citing Conference Paper
  • July 2022

... Introducing adversarial attacks to examine the performance under more adverse conditions makes the model more robust. Adversarial training is used for data augmentation and it makes models robust against attacks [136,137]. This method involves poisoning sample data during the training process, known as Robust Training, which has gained popularity across various domains. ...

Adversarial Training for Robust Insider Threat Detection
  • Citing Conference Paper
  • July 2022

... Any deviation from normal behavior results in suspicious activity. In this work, we follow the feature extraction adopted in [21] and reformulate in our FL setting, while Fig. 2 provides an overview of the non-IID feature generation followed in this work of distributed setting. As seen in Fig. 2, each client performs private feature extraction from the heterogeneous files to generate its private ITD dataset. ...

Anomaly Detection for Scenario-based Insider Activities using CGAN Augmented Data
  • Citing Conference Paper
  • October 2021

... The temporal characteristics of changes in greenhouse soil temperature are evident, and the internal climate of a greenhouse differs significantly from traditional meteorological forecasts. This is because both air and soil temperatures are greatly influenced by complex environmental factors and structural properties [20]. Relying solely on simple variables, such as temperature, for forecasting cannot adequately account for the temperature fluctuations caused by multiple interacting variables. ...

A survey on smart farming data, applications and techniques
  • Citing Article
  • June 2022

Computers in Industry