About
100
Publications
12,863
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
471
Citations
Introduction
Andri Ashfahani is a Ph.D. candidate in School of Computer Science and Engineering, Nanyang Technological University, Singapore. His research interest covers autonomous deep learning and data stream analytics for the complex environment. Currently, he is working on the incremental construction of a deep neural network.
Additional affiliations
May 2014 - March 2017
February 2013 - January 2014
Publications
Publications (100)
RFID technology has gained popularity to address
localization problem in the manufacturing shopfloor to track
the manufacturing object location to increase the production’s
efficiency. However, the data used for localization task is not
easy to analyze because it is generated from the non-stationary
environment. It also continuously arrive over tim...
The Denoising Autoencoder (DAE) enhances the flexibility of data stream method in exploiting unlabeled samples. Nonetheless, the feasibility of DAE for data stream analytic deserves in-depth study because it characterizes a fixed network capacity which cannot adapt to rapidly changing environments. Deep evolving denoising autoencoder (DEVDAN), is p...
Autonomous construction of deep neural network (DNNs) is desired
for data stream mining because it potentially offers two advantages:
proper model’s capacity and quick reaction to drift and shift.
While the self-organizing mechanism of DNNs remains an open issue,
this task is even more challenging to be developed for a standard
multi-layer DNNs tha...
The feasibility of existing data stream algorithms is often hindered by the weakly supervised condition of data streams. A self-evolving deep neural network, namely Parsimonious Network (ParsNet), is proposed as a solution to various weakly-supervised data stream problems. A self-labelling strategy with hedge (SLASH) is proposed in which its auto-c...
The long-standing challenge of building effective classification models for small and imbalanced datasets has seen little improvement since the creation of the Synthetic Minority Over-sampling Technique (SMOTE) over 20 years ago. Though GAN based models seem promising, there has been a lack of purpose built architectures for solving the aforementio...
Unsupervised continual learning remains a relatively uncharted territory in the existing literature because the vast majority of existing works call for unlimited access of ground truth incurring expensive labelling cost. Another issue lies in the problem of task boundaries and task IDs which must be known for model’s updates or model’s predictions...
A cross domain multistream classification is a challenging problem calling for fast domain adaptations to handle different but related streams in never-ending and rapidly changing environments. Notwithstanding that existing multistream classifiers assume no labelled samples in the target stream, they still incur expensive labelling cost since they...
A cross domain multistream classification is a challenging problem calling for fast domain adaptations to handle different but related streams in never-ending and rapidly changing environments. Notwithstanding that existing multistream classifiers assume no labeled samples in the target stream, they still incur expensive labeling costs since they r...
Designing a Convolutional Neural Networks (CNN) is a complex task and requires expert knowledge to optimize the performance and network architecture. In this paper, a novel data-driven approach is proposed to determine the architecture of CNN models. The proposed Autonomous Convolutional Neural Networks (AutoCNNThe executable code and original nume...
A deep clustering network is desired for data streams because of its aptitude in extracting natural features thus bypassing the laborious feature engineering step. While automatic construction of the deep networks in streaming environments remains an open issue, it is also hindered by the expensive labeling cost of data streams rendering the increa...
Data-driven quality monitoring is highly demanded in practise since it enables to relieve manual quality inspection of the product quality. Conventional data-driven quality monitoring is constrained by its offline characteristic thus being unable to handle streaming nature of sensory data and non-stationary environments of machine operations. Recen...
Unsupervised continual learning remains a relatively
uncharted territory in the existing literature
because the vast majority of existing works call
for unlimited access of ground truth incurring expensive
labelling cost. Another issue lies in the
problem of task boundaries and task IDs which
must be known for model’s updates or model’s predictions...
Pytorch implementation of Neural Network with Dynamically Evolved Capacity (NADINE) can be obtained from https://github.com/ContinualAL/NADINE_Pytorch
Pytorch implementation of Parsimonious Networks (ParsNet) can be obtained from https://github.com/ContinualAL/ParsNetPlus
Pytorch implementation of Autonomous Deep Learning (ADL) can be obtained from https://github.com/ContinualAL/ADL_Pytorch
Pytorch implementation of DEVDAN can be obtained from https://github.com/ContinualAL/DEVDAN_Pytorch
Unsupervised continual learning remains a relatively uncharted territory in the existing literature because the vast majority of existing works call for unlimited access of ground truth incurring expensive labelling cost. Another issue lies in the problem of task boundaries and task IDs which must be known for model's updates or model's predictions...
The common practice of quality monitoring in
industry relies on manual inspection well-known to be slow, errorprone
and operator-dependent. This issue raises strong demand
for automated real-time quality monitoring developed from datadriven
approaches thus alleviating from operator dependence
and adapting to various process uncertainties. Nonethele...
Designing an optimum Convolutional Neural Networks (CNN) is a complex task due to a large array of possible architectures and requires experience and in-depth knowledge of deep learning. This paper proposes Autonomous Convo-lutional Neural Networks (AutoCNN 1), a novel non-heuristic data-driven method to determine the CNN architecture for various c...
The common practice of quality monitoring in industry relies on manual inspection well-known to be slow, error-prone and operator-dependent. This issue raises strong demand for automated real-time quality monitoring developed from data-driven approaches thus alleviating from operator dependence and adapting to various process uncertainties. Nonethe...
Transferring knowledge across many streaming processes , i.e., multistream classification problem, advances the general transfer learning problem where it does not only require the domain adaptation strategy to align the source and target domains but also calls for scalable solution to handle the asynchronous drift problem between the source and ta...
A deep clustering network is desired for data
streams because of its aptitude in extracting natural features
thus bypassing the laborious feature engineering step. While
automatic construction of the deep networks in streaming en�vironments remains an open issue, it is also hindered by the
expensive labeling cost of data streams rendering the incre...
The numerical results of eT2QFNN from our repeated simulation
It shows the same results. link: https://www.researchgate.net/publication/324653944_An_Online_RFID_Localization_in_the_Manufacturing_Shopfloor
This document reports our recent runs of ATL codes shared here in researchgate. Our repeated simulation shows very similar results as reported in the paper (Reproducible). Note that random initialization method is applied here so every run will lead to different results. Here, we takes the average over five consecutive runs.
This document reports our recent runs of PARSNET codes shared here in researchgate. Our repeated simulation shows very similar results as reported in the paper (Reproducible). Note that random initialization method is applied here so every run will lead to different results. Here, we takes the average of five consecutive runs. Code is available her...
This document reports our recent runs of ADL codes shared here in researchgate. Our repeated simulation shows very similar results as reported in the paper (Reproducible). Note that random initialization method is applied here so every run will lead to different results. Here, we takes the average of five consecutive runs.
This document reports our recent runs of DEVDAN codes shared here in researchgate. Our repeated simulation shows very similar results as reported in the paper (Reproducible). Note that random initialization method is applied here so every run will lead to different results. Here, we takes the average of five consecutive runs.
This document reports our recent runs of NADINE codes shared here in researchgate. Our repeated simulation shows very similar results as reported in the paper (Reproducible). Note that random initialization method is applied here so every run will lead to different results. Here, we takes the average of five consecutive runs.
This is the original numerical results of the paper titled "Autonomous Deep Learning: Continual Learning Approach for Dynamic Environments". The code to reproduce these results can be downloaded from https://github.com/ContinualAL/ADL.
This document provides supplementary materials of our manuscript "Weakly Supervised Deep Learning Approach in Streaming Environments". It explains the key differences of ParsNet from DEVDAN, NADINE and ADL, visualizes ParsNet's learning performance, etc.
This document provides video demonstration for running ParNet matlab codes (m-file). It covers all aspects as indicated in the paper covering sporadic access of ground truth, infinitely delayed access of ground truth and ablation study.
This document provides Matlab codes of ParsNet (M-File) covering the three parts of our numerical study in the paper: sporadic access of ground truth, infinitely delayed access of ground truth and ablation study. Note that Xavier initialization method (random initialization method) is applied. In the paper, numerical results are obtained from the a...
This document contains the live code of our numerical results (HTML file ) as reported in the paper. If one is interested to obtain our raw numerical results, it can be provided by request. Code is available here: https://github.com/ContinualAL/ParsNet .
The feasibility of existing data stream algorithms is often hindered by the weakly supervised condition of data streams. A self-evolving deep neural network, namely Parsimonious Network (ParsNet), is proposed as a solution of various weakly-supervised data stream problems. A self-labelling strategy with hedge (SLASH) is proposed in which its auto-c...
This is the numerical results of ParsNet which are reported in the paper. The results are presented in the form of HTML for convenience purposes. If one is interested to get our raw numerical results as reported in the paper, we can provide it by request.
This video demonstrates the way to run ParsNet code.
This code implements ParsNet.
https://github.com/ContinualAL/ParsNet
Automatic detection of Attention Deficit Hyperactivity Disorder (ADHD) using resting-state - functional Magnetic Resonance Imaging (rs-fMRI) data is challenging due to the small number of samples, variations in acquisition technologies/techniques, imbalance in class distribution and high dimensionality of the data. In this paper, a Convolutional Ne...
Autonomous construction of deep neural network (DNNs) is desired for data streams because it potentially offers two advantages: proper model's capacity and quick reaction to drift and shift. While the self-organizing mechanism of DNNs remains an open issue, this task is even more challenging to be developed for standard multi-layer DNNs than that u...
The Denoising Autoencoder (DAE) enhances the flexibility of the data stream method in exploiting unlabeled samples. Nonetheless, the feasibility of DAE for data stream analytic deserves an in-depth study because it characterizes a fixed network capacity that cannot adapt to rapidly changing environments. Deep evolving denoising autoencoder (DEVDAN)...
These are the original numerical results where we study the effect of removing a component in order to provide additional insight into what makes NADINE performance. Specifically, we measure the effect of turning off the layer growing mechanism, turning off the node pruning mechanism, turning off the adaptive memory strategy and disabling the soft...
This is the reproduced numerical results of the paper titled "Autonomous Deep Learning: Continual Learning Approach for Dynamic Environments". The code to reproduce these results can be downloaded from https://github.com/ContinualAL/ADL.
Original numerical results of the paper titled "DEVDAN: Deep Evolving Denoising Autoencoder"
Original numerical results of the paper titled "DEVDAN: Deep Evolving Denoising Autoencoder"
Original numerical results of the paper titled "DEVDAN: Deep Evolving Denoising Autoencoder"
This is the numerical results of NADINE on regression problems. There are 3 regression problems used in our paper, i.e., household electric power, SP500, and condition monitoring. The result of the first problem is the original numerical result reported in the paper, whereas the numerical results of the last 2 problems are reproduced using the uplo...
This is the original numerical results of the paper titled "Automatic Construction of Multi-layer Perceptron Network from Streaming Examples". The results of classification problems are tabulated in Table 1 of the paper.
The dataset used in our experiment are available in: https://drive.google.com/drive/folders/1nbbiNAa6ZHIL9orNjQCQXuqyEXZ3bCdI?usp=sharing. You can find the code from https://github.com/ContinualAL/NADINE
This code is mFile implementation of our paper titled "DEVDAN: Deep Evolving Denoising Autoencoder". Inside the code, we have provided a list of the equation which may help the reader to find the implementation of every equation in the paper. All the datasets used in our experiments can be downloaded in this link: https://bit.ly/2mhtRsE. Note that...
This code is mFile implementation of our paper titled "Autonomous Deep Learning: Continual Learning Approach for Dynamic Environments". Inside the code, we have provided a list of the equation which may help the reader to find the implementation of every equation in the paper. All the datasets used in this paper can be downloaded in this link: http...
this folder contains the MATLAB code of our paper "Automatic Construction of Multi-layer Perceptron Network from Streaming Examples" published in CIKM 2019
The code can be downloaded here https://github.com/ContinualAL/ADL.
This explains several improvements in DEVDAN over ADL.
The code can be downloaded here https://github.com/ContinualAL/DEVDAN.
The feasibility of deep neural networks (DNNs) to address data stream problems still requires intensive study because of the static and offline nature of conventional deep learning approaches. A deep continual learning algorithm, namely autonomous deep learning (ADL), is proposed in this paper. Unlike traditional deep learning methods, ADL features...
With the recent explosion of data navigating in motion,
there is a growing research interest for analyzing streaming
data, and consequently, there are several recent works on data
stream analytics. However, exploring the potentials of traditional
recurrent neural network (RNN) in the context of streaming
data classification is still a little invest...
Radio Frequency Identification technology has gained popularity for cheap and easy deployment. In the realm of manufacturing shopfloor, it can be used to track the location of manufacturing objects to achieve better efficiency. The underlying challenge of localization lies in the nonstationary characteristics of manufacturing shopfloor which calls...
The feasibility of deep neural networks (DNNs) to address data stream problems still requires intensive study because of the static and offline nature of conventional deep learning approaches. A deep continual learning algorithm, namely autonomous deep learning (ADL), is proposed in this paper. Unlike traditional deep learning methods, ADL features...
Follow this link
https://github.com/ContinualAL/ADL.
The feasibility of deep neural networks (DNNs) to address data stream problems still requires intensive study because of the static and offline nature of conventional deep learning approaches. A deep continual learning algorithm, namely autonomous deep learning (ADL), is proposed in this paper. Unlike traditional deep learning methods, ADL features...
this code refers to all experiments in our paper "Autonomous Deep Learning: Continual Learning Approach for Dynamic Environments". It can be downloaded here https://github.com/ContinualAL/ADL.
The feasibility of deep neural networks (DNNs) to address data stream problems still requires intensive study because of the static and offline nature of conventional deep learning approaches. A deep continual learning algorithm, namely autonomous deep learning (ADL), is proposed in this paper. Unlike traditional deep learning methods, ADL features...
The generative learning phase of Autoencoder (AE) and its successor Denosing Autoencoder (DAE) enhances the flexibility of data stream method in exploiting unlabelled samples. Nonetheless, the feasibility of DAE for data stream analytic deserves in-depth study because it characterizes a fixed network capacity which cannot adapt to rapidly changing...
This is the code of evolving fuzzy neural network, namely evolving Type-2
Quantum Fuzzy Neural Network (eT2QFNN), which features an interval type-2
quantum fuzzy set with uncertain jump positions. The quantum fuzzy set
possesses a graded membership degree which enables better identification
of overlaps between classes. The eT2QFNN works fully in...
The paper (Autonomous Deep Learning: Incremental Learning of Denoising Autoencoder for Evolving Data Streams) was submitted to ICDM 2018 under this version.
RFID technology has gained popularity to address localization problem in the manufacturing shopfloor due to its affordability and easiness in deployment. This technology is used to track the manufacturing object location to increase the production's efficiency. However, the data used for localization task is not easy to analyze because it is genera...
Radio Frequency Identification technology has gained popularity for cheap and easy deployment. In the realm of manufacturing shopfloor, it can be used to track the location of manufacturing objects to achieve better efficiency. The underlying challenge of localization lies in the non-stationary characteristics of manufacturing shopfloor which calls...
Anti-lock braking system (ABS) is used on vehicles to keep the wheels unlocked in sudden break (inside braking) and minimalize the stop distance of the vehicle. The problem of it when sudden break is the wheels locked so the vehicle steering couldn't be controlled. The designed ABS system will be applied on ABS simulator using the electromagnetic b...
Utilizing additional devices for small signal stability enhancement of power systems is crucial over the last few years due to increasing number of demand, uncertainty of load and integrating intermittent power system based on renewable energy. Among the number of additional devices in power system, energy storage is becoming more popular in last f...
Nowadays, energy efficiency is an issue that is being faced by the industry. By increasing energy efficiency, production costs can be reduced to its lowest level. Terminal Transit Bahan Bakar Minyak Pertamina Wayame, Ambon as a part of the government committed to get actively involved in the implementation of the energy efficiency program. The inno...
A numerical investigation on the flow behavior past a circular cylinder at Reynolds number of Re = 1.0e6 - 8.4e6 is performed using Computational Fluid Dynamics (CFD) approach. The study is carried out using an overset grid method employing the two-equation eddy viscosity Menter SST turbulence model. For preliminary studies, simulations of two dime...
Stabilization problems for a mobile inverted pendulum are difficult because it is an unstable nonlinear system, and not all the states are available. This study demonstrated a
combination of constrained H∞ and loop transfer recovery (LTR) to control a mobile inverted pendulum. First, the linear model of a mobile inverted pendulum is derived via
li...
3-phase induction motors have been widely used in many industrial applications e.g. on the centrifugal machine for sugar manufacturing process. Centrifugal machine is used in liquid separation process. The motor should be able to track the reference to maintain the sugar quality. Therefore, the speed controller should be added to overcome this prob...
Stabilization problems for nonlinear systems are difficult, especially when not all the state are available. This study proposed a combination of constrained H∞ and loop transfer recovery (LTR) to control nonlinear systems via Takagi-Sugeno fuzzy model. First, the Takagi-Sugeno fuzzy model is employed to represent a nonlinear system. Next, based on...
Inverted pendulum system is a unstable and nonlinear system. It is commonly used to test the performance and eficuency of control method. This paper presents the design of a fuzzy controller for tracking of inverted pendulum system. The nonlinear model of inverted pendulum system is represented by using Takagi-Sugeno (T-S) fuzzy model. Tracking con...