Incremental Learning Model Research Articles

A More Robust Model to Answer Noisy Questions in KBQA

In practical applications, the raw input to a Knowledge Based Question Answering (KBQA) system may vary in forms, expressions, sources, etc. As a result, the actual input to the system may contain various errors caused by various noise in raw data and processes of transmission, transformation, translation, etc. As a result, it is significant to evaluate and enhance the robustness of a KBQA model to various noisy questions. In this paper, we generate 29 datasets of various noisy questions based on the original SimpleQuestions dataset to evaluate and enhance the robustness of a KBQA model, and propose a model which is more robust to various noisy questions. Compared with traditional methods, the main contribution in this paper is that we propose a method of generating datasets of different noisy questions to evaluate the robustness of a KBQA model, and propose a KBQA model which contains incremental learning and Mask Language Model (MLM) in the question answering process, so that our model is less affected by different kinds of noise in questions and achieves higher accuracies on datasets of different noisy questions, which shows its robustness. Experimental results show that our model achieves an average accuracy of 78.1% on these datasets and outperforms the baseline BERT-based model by an average margin of 5.0% with the similar training cost. In addition, further experiments show that our model is compatible with other pre-trained models such as ALBERT and ELECTRA.

Towards achieving lightweight intrusion detection systems in Internet of Things, the role of incremental machine learning: A systematic literature review

While the benefits of IoT cannot be overstated, its computational constraints make it challenging to deploy security methodologies that have been deployed in traditional computing systems. The benefits and computational constraints have made IoT systems attractive to cyber-attacks. One way to mitigate these attacks is to detect them. In this study, a Systematic Literature Review (SLR) has been conducted to analyze the role of incremental machine learning in achieving lightweight intrusion detection for IoT systems. The study analyzed existing incremental machine learning approaches used in designing intrusion detection systems for IoT ecosystems, emphasizing the incremental methods used in detecting intrusions, the datasets used to evaluate these methods, and how the method achieves lightweight status. The SLR outlined the contributions of each study, focusing on their strengths and gaps, the datasets used, and the incremental machine learning model used. This study revealed that incremental learning approaches in detecting intrusion in IoT systems are in their infant stage. Over 12 years, from 2010 to 2022, a total of twenty-one (21) studies were carried out in IDSs using incremental machine learning, with eight (8) studies carried out in IoT systems. In addition to reviewing the literature, we offer suggestions for improving existing solutions and achieving lightweight IDS for IoT systems. We also discussed some problems with making lightweight IDS for IoT systems and areas where more research could be done in the future.

ILL-IDS: An incremental lifetime learning IDS for VANETs

Vehicle Ad Hoc Network (VANET) is one of the most important approaches for intelligent vehicles to communicate under complex road conditions. However, as a VANET is working under wireless and complex conditions, it is under the threat of various malicious attacks. One efficient solution to counter those attacks is the Intrusion Detection System (IDS), which can detect intrusions to the VANET based on the statistical machine learning method. However, most IDS solutions have a problem that they can only detect the attacks which have already known when the IDS is designed/trained. For unknown attacks, most IDSs become inefficient. A blockchain-based Lifetime Learning IDS (LL-IDS) framework has been designed to address this problem. It applies a blockchain to store uncertain data that an IDS cannot decide and is highly likely to be a new type of attacks. With the help of traditional security agencies such as universities, these uncertain data can be labelled to train the IDS model. The incremental learning models are shown to have a great potential in this scenario. In this paper, we introduces a novel IDS named as incremental IDS based on LL-IDS. Numerical experiments show that the computational time consumption and web payload could be decreased by applying the ILL-IDS to a public VANET dataset.

An Incremental Learning Model for Mobile Robot: From Short-Term Memory to Long-Term Memory

During the environmental cognition, how to realize the efficient incremental learning of a mobile robot is a great challenge. The existing methods suffer from the low efficiency. This article proposes a novel methodology to address it. A memory model inspired by the human brains is constructed to realize the transmission of short-term memory (STM) to long-term memory (LTM) in the offline states, thus realize the efficient incremental learning of the mobile robot. Concretely, during the online process, when the sensory information is input to the developmental network (DN), similarity between the new input information and the knowledge memorized in the DN is calculated first. If the similarity is larger than the threshold, this input is transmitted to motor layer and determine the optimal decision. Otherwise, the sensory input will be temporarily stored in a neuron in the STM by an evaluation function. During the offline states, a self-triggering mechanism is designed to trigger the DN to work again. Then, the lateral excitation of the internal neurons is designed to fire more neurons to memorize the knowledge transferred from the STM. Then, the synaptic weights of the new fire neurons are updated, and the STM becomes the LTM, thus realizing the incremental learning. In the following task, if the robot encounters a similar scenario, it can make a quick decision, based on the knowledge learned during the offline states. This continuous learning pattern reduces the training samples used in the network, hence enhancing its learning efficiency. Most importantly, this methodology makes the robot continuously improve its intelligence through the incremental learning, even in off-line states. Extensive simulation and experiment results of the mobile robot navigation demonstrate its potential.

Discrimination Correction and Balance for Class-Incremental Learning

Catastrophic forgetting is the key role in incremental learning, since the model gains poor quality for old classes than new classes with continuously incoming tasks and less storage. In order to solve the balance between model plasticity and stability, this paper proposes a novel class incremental learning method DCBIL, which adopts discrimination correction and balance to avoid model bias. First, three components Conservative, Balancer and Opener with identical structure are designed. Then, Conservative and Opener have a bias to old and new classes respectively. And Balancer modifies the fully connected layer and replaces the base model for the next training. Moreover, a stochastic perturbation probability synergy is incorporated for final output fusion. Experimental evaluations on Mnist, Cifar-10, Cifar-100 and TinyImageNet demonstrate the effectiveness of DCBIL. DCBIL is portable for it can be used as a plug-in to strengthen the performance of existing incremental learning model.

Incremental learning-inspired mating restriction strategy for Evolutionary Multiobjective Optimization

The prior knowledge from the problem property can boost the evolutionary multiobjective optimization (EMO). The existing machine learning model for knowledge mining in the EMO has led to enhanced performance on multiobjective optimization problems with complicated Pareto fronts (MOPs-cPF), but the high computational cost resulting from model training should not be taken as a natural expense. To overcome this drawback, this paper proposes an incremental learning-inspired mating restriction strategy (ILMR) for solving MOP-cPF efficiently. In ILMR, a mating restriction is implemented based on an incremental learning model that establishes the mating pool for each solution in the population and incrementally updates as the population evolves. Specifically, it consists of two interdependent parts, i.e., a learning module and a forgetting module. In one evolutionary cycle, the learning module is used to learn new knowledge from the high-quality offspring solutions, while the forgetting module is utilized to remove the information provided by relatively poor solutions in the population. Moreover, a multiobjective evolutionary algorithm with ILMR, named MEILM, is proposed and compared with six state-of-the-art algorithms on a variety of MOP-cPF. The experimental results show that there are significant improvements benefitting from the proposed mating restriction strategy.

A real-time prediction method for tunnel boring machine cutter-head torque using bidirectional long short-term memory networks optimized by multi-algorithm

Based on data from the Jilin Water Diversion Tunnels from the Songhua River (China), an improved and real-time prediction method optimized by multi-algorithm for tunnel boring machine (TBM) cutter-head torque is presented. Firstly, a function excluding invalid and abnormal data is established to distinguish TBM operating state, and a feature selection method based on the SelectKBest algorithm is proposed. Accordingly, ten features that are most closely related to the cutter-head torque are selected as input variables, which, in descending order of influence, include the sum of motor torque, cutter-head power, sum of motor power, sum of motor current, advance rate, cutter-head pressure, total thrust force, penetration rate, cutter-head rotational velocity, and field penetration index. Secondly, a real-time cutter-head torque prediction model's structure is developed, based on the bidirectional long short-term memory (BLSTM) network integrating the dropout algorithm to prevent overfitting. Then, an algorithm to optimize hyperparameters of model based on Bayesian and cross-validation is proposed. Early stopping and checkpoint algorithms are integrated to optimize the training process. Finally, a BLSTM-based real-time cutter-head torque prediction model is developed, which fully utilizes the previous time-series tunneling information. The mean absolute percentage error ( MAPE ) of the model in the verification section is 7.3%, implying that the presented model is suitable for real-time cutter-head torque prediction. Furthermore, an incremental learning method based on the above base model is introduced to improve the adaptability of the model during the TBM tunneling. Comparison of the prediction performance between the base and incremental learning models in the same tunneling section shows that: (1) the MAPE of the predicted results of the BLSTM-based real-time cutter-head torque prediction model remains below 10%, and both the coefficient of determination ( R 2 ) and correlation coefficient ( r ) between measured and predicted values exceed 0.95; and (2) the incremental learning method is suitable for real-time cutter-head torque prediction and can effectively improve the prediction accuracy and generalization capacity of the model during the excavation process. • A torque prediction model based on BLSTM and multi-algorithm fusion is developed. • The Dropout algorithm is integrated to prevent overfitting. • Bayesian and Cross-Validation based hyperparameter optimization method is presented. • EarlyStopping and ModelCheckPoint are integrated to optimize the training process. • An incremental learning method is proposed to improve the model's adaptability.

An incremental learning approach to prediction models of SEIRD variables in the context of the COVID-19 pandemic.

Several works have proposed predictive models of the SEIRD (Susceptible, Exposed, Infected, Recovered, and Dead) variables to characterize the pandemic of COVID-19. One of the challenges of these models is to be able to follow the dynamics of the disease to make more precise predictions. In this paper, we propose an approach based on incremental learning to build predictive models of the SEIRD variables for the COVID-19 pandemic. Our incremental learning approach is a dynamic ensemble method based on a bagging scheme that allows the addition of new models or the updating of incremental models. The article proposes an incremental learning architecture composed of two components. The first component carries out an analysis of the interdependencies of the SEIRD variables and the second component is an incremental learning model that builds/updates the predictive models. The paper analyses the quality of the predictive models of our incremental learning approach using data of the COVID-19 from Colombia, and shows interesting results about the predictions of the SEIRD variables. These results are compared with an incremental learning approach based on random forests.

A multiobjective prediction model with incremental learning ability by developing a multi-source filter neural network for the electrolytic aluminium process

Improving current efficiency and reducing energy consumption are two important technical goals of the electrolytic aluminum process (EAP). However, because the process involves complex noise characteristics (i.e., unknown types, redundant distributions and variable forms), it is very difficult to accurately develop a multiobjective prediction model. To overcome this problem, in this paper, a novel framework of multiobjective incremental learning based on a multi-source filter neural network (MSFNN) is presented. The proposed framework first presents a “multi-source filter” (MSF) technique that utilizes the mean and variance in the unscented Kalman filter (UKF) to guide the importance function of the particle filter (PF) based on a density kernel estimation method. Then, the MSF is embedded in the mutated neural network to adjust weights in real time. Third, weights are calculated and normalized by a modified importance function, which is the basis for further optimizing a secondary sampling based on sampling importance resampling (SIR). Finally, the incremental learning model with two objectives (i.e., process power consumption and current efficiency) based on the MSFNN in the EAP is established. The presented framework has been verified by the real-world EAP and some closely related methods. All test results indicate that the MSFNN’s relative prediction errors of the above two objectives are controlled within 0.51% and 0.38%, respectively and prove that MSFNN has significant competitive advantages over other recent filtering network models. Successfully establishment of the proposed framework provides a model foundation for multiobjective optimization problems in the EAP.

Interpretable Short-Term Electrical Load Forecasting Scheme Using Cubist

Daily peak load forecasting (DPLF) and total daily load forecasting (TDLF) are essential for optimal power system operation from one day to one week later. This study develops a Cubist-based incremental learning model to perform accurate and interpretable DPLF and TDLF. To this end, we employ time-series cross-validation to effectively reflect recent electrical load trends and patterns when constructing the model. We also analyze variable importance to identify the most crucial factors in the Cubist model. In the experiments, we used two publicly available building datasets and three educational building cluster datasets. The results showed that the proposed model yielded averages of 7.77 and 10.06 in mean absolute percentage error and coefficient of variation of the root mean square error, respectively. We also confirmed that temperature and holiday information are significant external factors, and electrical loads one day and one week ago are significant internal factors.

Online multimedia traffic classification from the QoS perspective using deep learning

In the next generation of communication systems, data traffic is expected to increase dramatically and continuously. Particularly for multimedia traffic, it has a dominant share in the increasing traffic. Therefore, there is an urgent need to develop an effective and accurate scheme to achieve online and automatic traffic management. To this end, this paper proposes an online multimedia traffic classification framework based on a Convolutional Neural Network (CNN), capable of conducting fast and early classification as well as class incremental learning. First, the sliding window technique is applied to capture the flow slices for further feature extraction. Then, the 3-dimensional flow representation is extracted based on the probability distribution function. After that, according to the specific structure of features, a deeply adapted structure of CNN is devised to better learn the knowledge from the representation. Besides, to better support the addition of new services, a class incremental learning model is developed with the techniques of knowledge distillation and bias correction to achieve continuous learning without retraining from scratch. Our experimental results reveal that the proposed method achieves faster and more accurate traffic classification compared with the state-of-the-art. Additionally, the deployed scheme using incremental learning achieves drops by about 50% in both time and memory consumptions compared with existing methods, while guaranteeing the accurate classification after adding new classes.

Incremental Learning Model for Load Forecasting without Training Sample

This article presents hourly load forecasting by using an incremental learning model called Online Sequential Extreme Learning Machine (OS-ELM), which can learn and adapt automatically according to new arrival input. However, the use of OS-ELM requires a sufficient amount of initial training sample data, which makes OS-ELM inoperable if sufficiently accurate sample data cannot be obtained. To solve this problem, a synthesis of the initial training sample is proposed. The synthesis of the initial sample is achieved by taking the first data received at the start of working and adding random noises to that data to create new and sufficient samples. Then the synthesis samples are used to initial train the OS-ELM. This proposed method is compared with Fully Online Extreme Learning Machine (FOS-ELM), which is an incremental learning model that also does not require the initial training samples. Both the proposed method and FOS-ELM are used for hourly load forecasting from the Hourly Energy Consumption dataset. Experiments have shown that the proposed method with a wide range of noise levels, can forecast hourly load more accurately than the FOS-ELM.

Flow-Based Reinforcement Learning

This paper presents a novel Flow-based reinforcement learning strategy to model agent systems that can adapt to complex and dynamic problem environments by incrementally mastering their skills. It is inspired by the psychological notion of Flow that describes the optimal mental state experienced by an individual when they are fully immersed in a task and find it intrinsically rewarding to engage with. The proposed model presents an algorithm to describe the Flow experience such that agents can be trained through finer distinctions to the challenges across training time to maintain them in the Flow zone. In contrast to the traditional and incremental learning approaches that suffer from limitations associated with overfitting, the Flow-based model drives agent behaviours not simply through external goals but also through intrinsic curiosity to improve their skills and thus the performance levels. Experimental evaluations are conducted across two simulation environments on a maze navigation task and a reward collection task with comparisons against a generic reinforcement learning model and an incremental reinforcement learning model. The results reveal that these two models are prone to overfit under different design decisions and loose the ability to perform in dynamic variations of the tasks in varying degrees. Conversely, the proposed Flow-based model is capable of achieving near optimal solutions with random environmental factors, appropriately utilising the previously learned knowledge to identify robust solutions to complex problems.

IDUDL: Incremental Double Unsupervised Deep Learning Model for Marine Aquaculture SAR Images Segmentation

Marine aquaculture is an important natural resource exploration that requires rational planning to avoid environmental damage. Synthetic Aperture Radar (SAR) images are essential in remote sensing to monitor the marine ecological environment. Unsupervised methods provide an adequate solution to avoid the cost of training sample collection. However, unsupervised methods often struggle to discover effective semantic information and incrementally utilize newly acquired data. To address these challenges, this article presents an incremental double unsupervised deep learning (IDUDL) model, which is specially designed to characterize unlabeled marine aquaculture and achieve the results semantically. Based on the idea of alternately generating and updating pseudo-labels, the proposed IDUDL model defines the double neural networks comprised of the feature extraction network (FEN) and the fully convolutional semantic segmentation network (FCSSN). A patch estimation (PE) is proposed to generate pseudo-labels with aquaculture semantic information based on the features extracted by the FEN network. Then, the aquaculture extraction results are obtained by the FCSSN with generated pseudo-labels. After that, the pseudo-labels and extraction results are updated in turns until the pseudo-labels are stable. In addition, due to the unique structure of double neural networks, newly acquired marine aquaculture SAR images can also be added to the pre-trained FCSSN and followed pseudo-labels updated based on the FEN and PE part, which can achieve new data incremental learning without retrained the whole IDUDL model. Experiments demonstrate the effectiveness of the proposed approach based on two different ways of the marine aquaculture including raft and cage types, which consist of GaoFen-3 (GF-3) and RADARSAT-2 SAR images from the Dalian and Ningde areas, respectively. The incremental experiments are also designed to verify the generalization of the IDUDL model for new obtained marine aquaculture SAR images. The code of this work will be available at https://github. com/fjc1575/Marine-Aquaculture/tree/main/IDUDL for the sake of reproducibility.

Active Inference Integrated With Imitation Learning for Autonomous Driving

Classical imitation learning methods suffer substantially from the learning hierarchical policies when the imitative agent faces an unobserved state by the expert agent. To address these drawbacks, we propose an online active learning through active inference approach that encodes the expert’s demonstrations based on observation-action to improve the learner’s future motion prediction. For this purpose, we provide a switching Dynamic Bayesian Network based on the dynamic interaction between the expert agent and another object in its surrounding as a reference model, which we exploit to initialize an incremental probabilistic learning model. This learning model grows and matures based on the free-energy formulation and message passing of active inference dynamically at discrete and continuous levels in an online active learning phase. In this scheme, generalized states of the learning world are represented as distance-vector, where it is the learner’s observation concerning its interaction with a moving object. Considering the distance vector entail intentions, it enables action prediction evaluation in a prospective sense. We illustrate these points using simulations of driving intelligent agents. The learning agent is trained by using long-term predictions from the generative learning model to reproduce the expert’s motion while learning how to select a suitable action through new experiences. Our results affirm that a Dynamic Bayesian optimal approach provides a principled framework and outperforms conventional reinforcement learning methods. Furthermore, it endorses the general formulation of action prediction as active inference.

LIL: Lightweight Incremental Learning Approach Through Feature Transfer for Remote Sensing Image Scene Classification

Existing deep learning models usually assume that all data obeys independent identically distribution, which is unreasonable in remote sensing. Due to the differences in camera parameters, spectral ranges, resolutions, and so on, the images acquired by remote sensing sensors may be greatly diverse, causing models to face catastrophic forgetting when they are trained on new data only. Thus, incremental learning is introduced. An ideal incremental learning model should be expanded as the number of tasks increases, so as to have enough ability to adapt to the changes in data. However, existing approaches normally expand heavy modules for each task, making the holistic models cumbersome. In this article, a lightweight incremental learning approach (LIL) is proposed for remote sensing image scene classification. We replace the role of the feature extractor with extracting features of a single task instead of task-sharing features of all tasks to lighten the backbone. In addition, we propose a light feature transfer module (FTM) to realize the alignment of data distributions between different tasks in the feature domain. Furthermore, dual-constraint loss with knowledge distillation and adversarial learning is introduced to promote the mapping and alignment of data distributions at both the feature level and the semantic level. In LIL, only a tiny FTM and a classifier are added to the model when the model learns a new task. Experimental results show that our approach with a small number of parameters outperforms state-of-the-art approaches for incremental learning on both a single dataset and a sequence of multiple datasets.

Deep Incremental Learning for Personalized Human Activity Recognition on Edge Devices

Tracking human daily activities is a useful functionality supported in many applications, especially with the pervasive use of wearable devices. State-of-the-art human activity recognition (HAR) uses machine or deep learning techniques to identify activities based on sensor readings. However, these models represent patterns from standardized experiment setups, with limited diversity when it comes to the individuals involved in data collection. This leads to limited success of HAR in real deployment scenarios, where individuals perform the same activity in different ways. Training models from scratch on real-time data streams is challenging due to the computational complexity of machine and deep learning architectures. In this article, we propose an incremental learning model for HAR that tweaks a deep learning model pretrained on a standardized HAR dataset and incrementally trains on newly generated individuals personalized data on their personal devices. The proposed solution promotes the preservation of data privacy, improves the model performance in terms of accuracy and efficiency without having to retrain from scratch, and tweaks the model according to personalized activity patterns. Extensive experiments show improvement of the base model’s accuracy up to 19% after incrementally training the model on filtered users’ datasets for the standing, walking, and running activities.

A-iLearn: An adaptive incremental learning model for spoof fingerprint detection

Incremental learning enables the learner to accommodate new knowledge without retraining the existing model. It is a challenging task that requires learning from new data and preserving the knowledge extracted from the previously accessed data. This challenge is known as the stability-plasticity dilemma. We propose A-iLearn, a generic model for incremental learning which overcomes the stability-plasticity dilemma by carefully integrating the ensemble of base classifiers trained on new data with the current ensemble without retraining the model from scratch using entire data. We demonstrate the efficacy of the proposed A-iLearn model on spoof fingerprint detection application. One of the significant challenges associated with spoof fingerprint detection is the performance drop on spoofs generated using new fabrication materials. A-iLearn is an adaptive incremental learning model that adapts to the features of the “live” and “spoof” fingerprint images and efficiently recognizes the new spoof fingerprints and the known spoof fingerprints when the new data is available. To the best of our knowledge, A-iLearn is the first attempt in incremental learning algorithms that adapts to the properties of data for generating a diverse ensemble of base classifiers. From the experiments conducted on standard high-dimensional datasets LivDet 2011, LivDet 2013 and LivDet 2015, we show that the performance gain on new fake materials is significantly high. On average, we achieve 49.57% improvement in accuracy between the consecutive learning phases.

A defect detection system for wire arc additive manufacturing using incremental learning

In more recent times, research on various aspects of the Wire Arc Additive Manufacturing (WAAM) process has been conducted, and efforts into monitoring the WAAM process for defect identification have increased. Rapid and reliable monitoring of the WAAM process is a key development for the technology as a whole, as it will enable components produced by the process to be qualified to relevant standards and hence be deemed fit for use in applications such as those found in the aerospace or naval sectors. Intelligent algorithms provide inbuilt advantages in processing and analysing data, especially for the large data sets generated during the long manufacturing cycles. Interdisciplinary engineering (IDE) furnishes a concept integrating computer science and industrial system manufacturing engineering together to treat large amounts of process monitoring data. In this work, a WAAM process monitoring and defect detection system integrating intelligent algorithms is presented. The system monitors welding arc current and voltage signals produced by the WAAM process and makes use of a support vector machine (SVM) learning method to identify disturbances to the welding signal which indicate the presence of potential defects. The incremental machine learning models developed in this work are trained via statistical feature analysis of the welding signals and a novel quality metric that improves detection rates is also presented. The incremental learning approach provides an efficient means of detecting welding-based defects, as it does not require large quantities of data to be trained to an operational level (addressing a major drawback of other machine learning methods). A case study is presented to validate the developed system, results show that it was able to detect a set of defects with a success rate greater than 90% F1-score.

Incremental human action recognition with dual memory

Incremental learning is a topic of great interest in the current state of machine learning research. Real-world problems often require a classifier to incorporate new knowledge while preserving what was learned before. One of the most challenging problems in computer vision is Human Action Recognition (HAR) in videos. However, most of the existing works approach HAR from a non-incremental point of view. This work proposes a framework for performing HAR in the incremental learning scenario called Incremental Human Action Recognition with Dual Memory (IHAR-DM). IHAR-DM contains three main components: a 3D convolutional neural network for capturing Spatio-temporal features; a Triplet Network to perform metric learning; and the dual-memory Extreme Value Machine, which is introduced in this work. The proposed method is compared with 10 other state-of-the-art incremental learning models. We propose five experimental settings containing different numbers of tasks and classes using two widely known HAR datasets: UCF-101 and HMDB51. Our results show superior performance in terms of Normalized Mutual Information (NMI) and Inter-task Intransigence (ITI), which is a new metric proposed in this work. Overall results show the feasibility of the proposal for real HAR problems, which mostly present the requirements imposed by incremental learning.