Pre-trained language models with limited data for intent classification
- Authors: Kasthuriarachchy, Buddhika , Chetty, Madhu , Karmakar, Gour , Walls, Darren
- Date: 2020
- Type: Text , Conference proceedings , Conference paper
- Relation: 2020 International Joint Conference on Neural Networks, IJCNN 2020
- Full Text: false
- Reviewed:
- Description: Intent analysis is capturing the attention of both the industry and academia due to its commercial and noncommercial significance. The rapid growth of unstructured data of micro-blogging platforms, such as Twitter and Facebook, are amongst the important sources for intent analysis. However, the social media data are often noisy and diverse, thus making the task very challenging. Further, the intent analysis frequently suffers from lack of sufficient data because the labeled datasets are often manually annotated. Recently, BERT (Bidirectional Encoder Representation from Transformers), a state-of-the-art language representation model, has attracted attention for accurate language modelling. In this paper, we investigate the application of BERT for its suitability for intent analysis. We study the fine-tuning of the BERT model through inductive transfer learning and investigate methods to overcome the challenges due to limited data availability by proposing a novel semantic data augmentation approach. This technique generates synthetic sentences while preserving the label-compatibility using the semantic meaning of the sentences, to improve the intent classification accuracy. Thus, based on the considerations for finetuning and data augmentation, a systematic and novel step-bystep methodology is presented for applying the linguistic model BERT for intent classification with limited data available. Our results show that the pre-trained language can be effectively used with noisy social media data to achieve state-of-the-art accuracy in intent analysis under low labeled-data regime. Moreover, our results also confirm that the proposed text augmentation technique is effective in eliminating noisy synthetic sentences, thereby achieving further performance improvements. © 2020 IEEE.
Assessing transformer oil quality using deep convolutional networks
- Authors: Alam, Mohammad , Karmakar, Gour , Islam, Syed , Kamruzzaman, Joarder , Chetty, Madhu , Lim, Suryani , Appuhamillage, Gayan , Chattopadhyay, Gopi , Wilcox, Steve , Verheyen, Vincent
- Date: 2019
- Type: Text , Conference proceedings , Conference paper
- Relation: 29th Australasian Universities Power Engineering Conference, AUPEC 2019
- Full Text:
- Reviewed:
- Description: Electrical power grids comprise a significantly large number of transformers that interconnect power generation, transmission and distribution. These transformers having different MVA ratings are critical assets that require proper maintenance to provide long and uninterrupted electrical service. The mineral oil, an essential component of any transformer, not only provides cooling but also acts as an insulating medium within the transformer. The quality and the key dissolved properties of insulating mineral oil for the transformer are critical with its proper and reliable operation. However, traditional chemical diagnostic methods are expensive and time-consuming. A transformer oil image analysis approach, based on the entropy value of oil, which is inexpensive, effective and quick. However, the inability of entropy to estimate the vital transformer oil properties such as equivalent age, Neutralization Number (NN), dissipation factor (tanδ) and power factor (PF); and many intuitively derived constants usage limit its estimation accuracy. To address this issue, in this paper, we introduce an innovative transformer oil analysis using two deep convolutional learning techniques such as Convolutional Neural Network (ConvNet) and Residual Neural Network (ResNet). These two deep neural networks are chosen for this project as they have superior performance in computer vision. After estimating the equivalent aging year of transformer oil from its image by our proposed method, NN, tanδ and PF are computed using that estimated age. Our deep learning based techniques can accurately predict the transformer oil equivalent age, leading to calculate NN, tanδ and PF more accurately. The root means square error of estimated equivalent age produced by entropy, ConvNet and ResNet based methods are 0.718, 0.122 and 0.065, respectively. ConvNet and ResNet based methods have reduced the error of the oil age estimation by 83% and 91%, respectively compared to that of the entropy method. Our proposed oil image analysis can calculate the equivalent age that is very close to the actual age for all images used in the experiment. © 2019 IEEE.
- Description: E1
Measuring trustworthiness of IoT image sensor data using other sensors' complementary multimodal data
- Authors: Islam, Mohammad , Karmakar, Gour , Kamruzzaman, Joarder , Murshed, Manzur
- Date: 2019
- Type: Text , Conference proceedings , Conference paper
- Relation: 18th IEEE International Conference on Trust, Security and Privacy in Computing and Communications/13th IEEE International Conference on Big Data Science and Engineering, TrustCom/BigDataSE 2019 p. 775-780
- Full Text: false
- Reviewed:
- Description: Trust of image sensor data is becoming increasingly important as the Internet of Things (IoT) applications grow from home appliances to surveillance. Up to our knowledge, there exists only one work in literature that estimates trustworthiness of digital images applied to forensic applications, based on a machine learning technique. The efficacy of this technique is heavily dependent on availability of an appropriate training set and adequate variation of IoT sensor data with noise, interference and environmental condition, but availability of such data cannot be assured always. Therefore, to overcome this limitation, a robust method capable of estimating trustworthy measure with high accuracy is needed. Lowering cost of sensors allow many IoT applications to use multiple types of sensors to observe the same event. In such cases, complementary multimodal data of one sensor can be exploited to measure trust level of another sensor data. In this paper, for the first time, we introduce a completely new approach to estimate the trustworthiness of an image sensor data using another sensor's numerical data. We develop a theoretical model using the Dempster-Shafer theory (DST) framework. The efficacy of the proposed model in estimating trust level of an image sensor data is analyzed by observing a fire event using IoT image and temperature sensor data in a residential setup under different scenarios. The proposed model produces highly accurate trust level in all scenarios with authentic and forged image data. © 2019 IEEE.
- Description: E1
Trusted autonomous vehicle : measuring trust using on-board unit data
- Authors: Chowdhury, Abdullahi , Karmakar, Gour , Kamruzzaman, Joarder
- Date: 2019
- Type: Text , Conference proceedings , Conference paper
- Relation: 18th IEEE International Conference on Trust, Security and Privacy in Computing and Communications/13th IEEE International Conference on Big Data Science and Engineering, TrustCom/BigDataSE 2019 p. 787-792
- Full Text: false
- Reviewed:
- Description: Vehicular Ad-hoc Networks (VANETs) play an essential role in ensuring safe, reliable and faster transportation with the help of an Intelligent Transportation system. The trustworthiness of vehicles in VANETs is extremely important to ensure the authenticity of messages and traffic information transmitted in extremely dynamic topographical conditions where vehicles move at high speed. False or misleading information may cause substantial traffic congestions, road accidents and may even cost lives. Many approaches exist in literature to measure the trustworthiness of GPS data and messages of an Autonomous Vehicle (AV). To the best of our knowledge, they have not considered the trustworthiness of other On-Board Unit (OBU) components of an AV, along with GPS data and transmitted messages, though they have a substantial relevance in overall vehicle trust measurement. In this paper, we introduce a novel model to measure the overall trustworthiness of an AV considering four different OBU components additionally. The performance of the proposed method is evaluated with a traffic simulation model developed by Simulation of Urban Mobility (SUMO) using realistic traffic data and considering different levels of uncertainty. © 2019 IEEE.
- Description: E1
Passive detection of splicing and copy-move attacks in image forgery
- Authors: Islam, Mohammad , Kamruzzaman, Joarder , Karmakar, Gour , Murshed, Manzur , Kahandawa, Gayan
- Date: 2018
- Type: Text , Conference proceedings , Conference paper
- Relation: 25th International Conference on Neural Information Processing, ICONIP 2018; Siem Reap, Cambodia; 13th-16th December 2018; published in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Vol. 11304 LNCS, p. 555-567
- Full Text:
- Reviewed:
- Description: Internet of Things (IoT) image sensors for surveillance and monitoring, digital cameras, smart phones and social media generate huge volume of digital images every day. Image splicing and copy-move attacks are the most common types of image forgery that can be done very easily using modern photo editing software. Recently, digital forensics has drawn much attention to detect such tampering on images. In this paper, we introduce a novel feature extraction technique, namely Sum of Relevant Inter-Cell Values (SRIV) using which we propose a passive (blind) image forgery detection method based on Discrete Cosine Transformation (DCT) and Local Binary Pattern (LBP). First, the input image is divided into non-overlapping blocks and 2D block DCT is applied to capture the changes of a tampered image in the frequency domain. Then LBP operator is applied to enhance the local changes among the neighbouring DCT coefficients, magnifying the changes in high frequency components resulting from splicing and copy-move attacks. The resulting LBP image is again divided into non-overlapping blocks. Finally, SRIV is applied on the LBP image blocks to extract features which are then fed into a Support Vector Machine (SVM) classifier to identify forged images from authentic ones. Extensive experiment on four well-known benchmark datasets of tampered images reveal the superiority of our method over recent state-of-the-art methods.
Dynamically controlling exterior and interior window coverings through IoT for environmental friendly smart homes
- Authors: Karmakar, Gour , Roy, Soma , Chattopadhyay, Gopinath , Xiao, Zhigang
- Date: 2017
- Type: Text , Conference proceedings , Conference paper
- Relation: 2017 IEEE International Conference on Mechatronics, ICM 2017; Gippsland, Australia; 13th-15th February 2017 p. 487-491
- Full Text: false
- Reviewed:
- Description: Energy saving using smart home is of paramount importance to reduce heating and cooling energy consumption, and promote sustainable environment. Awnings and blinds have exhibited their effectiveness to reduce heating gain in summer and cooling loss in winter, respectively. Awnings are more effective to reduce heat gain in summer than blinds, while the opposite is true in winter. There exist many approaches in the current literature to remotely control flat curtains and blinds. However, up to our knowledge, no automatic technique is available in the literature, which can dynamically control the orientation of an exterior covering so that it can act like a blind in winter and an awning in summer. In this paper, we propose an automatic on-demand system to control the orientation and size of such exterior covering, and the turning air conditioners, heaters and lights on and off considering the rate of change of room temperature, and its lighting condition. We also discuss the properties and design of such exterior covering. A simulation model was developed to analysis the performance of our approach in terms of energy savings both in summer and winter. © 2017 IEEE.
- Description: Proceedings - 2017 IEEE International Conference on Mechatronics, ICM 2017
Exploiting evolving trust relationships in the modelling of opinion formation dynamics in online social networks
- Authors: Das, Rajkumar , Kamruzzaman, Joarder , Karmakar, Gour
- Date: 2017
- Type: Text , Conference proceedings , Conference paper
- Relation: 31st IEEE International Conference on Advanced Information Networking and Applications, AINA 2017; Taipei, Taiwan; 27th-29th March 2017 p. 872-879
- Full Text: false
- Reviewed:
- Description: Mass participation of the members of a society in discussions to resolve issues related to a topic leads to forming public opinion. The timeline of the underlying dynamics goes through several distinguishable phases, and experiences transition from one to another. After initiated by concerned individuals, it draws active attention from almost everyone, and with time progression, people's participation starts declining as the issues are resolved or lost attraction. The existing works in the literature to capture the opinion formation process pay attention to model the dynamics in its active phase and thus ignore the other phases and the corresponding phase transitions. Trust relationships among the participants dynamically shape their interactions in different stages of the dynamics. Existing works fail to incorporate trust in defining the extent of influence one has on others, as they define the social relationships in the opinion space. To address this issue, we adopt simulated annealing to model the transitional behaviour of the dynamics, and then, amalgamate peoples relationships in the trust space with that in the opinion space to define the meta-heuristics of the algorithm for capturing the dynamical properties of the process. Finally, through simulation, we observe that our model is insightful in representing peoples' evolving behaviour in the different stages of opinion formation process, and consequently, can capture the various properties of the steady-state outcomes of the dynamics. © 2017 IEEE.
- Description: Proceedings - International Conference on Advanced Information Networking and Applications, AINA
Who are convincing? An experience based opinion formation dynamics in online social networks
- Authors: Das, Rajkumar , Kamruzzaman, Joarder , Karmakar, Gour
- Date: 2016
- Type: Text , Conference proceedings , Conference paper
- Relation: 30th European Simulation and Modelling Conference, ESM 2016; Las Palmas, Spain; 26th-28th October 2016 p. 167-173
- Full Text: false
- Reviewed:
- Description: Online social network (OSN) is one of the major platforms where our opinions are formed now-a-days and increasing so. Opinion formation dynamics captures the ways public opinions are formed, mainly from two different sources, (i) neighbours' opinions, (ii) external opinions from sources other than the neighbours. In this paper, we formulate an opinion formation model by considering two very important factors, that were ignored or a very little explored in the literature. First, we model the convincing power of the opinions encountered from the two sources. Second, we incorporate the experience of users' previous interactions with the two opinion sources. The problem is formulated as an agent based model where each member of an OSN is represented with an agent and their relationships with a graph. Finally through simulation, we create various scenarios, and apply our model to observe the steady state outcomes of the dynamics. This helps us to study the nature of the public opinions under various influences of our model parameters.
- Description: European Simulation and Modelling Conference 2016, ESM 2016