Deep neural networks (DNNs) often suffer from "catastrophic forgetting" during incremental learning (IL) --- an abrupt degradation of performance on the original set of classes when the training objective is adapted to a newly added set of classes. Existing IL approaches tend to produce a model that is biased towards either the old classes or new classes, unless with the help of exemplars of the old data. To address this issue, we propose a class-incremental learning paradigm called Deep Model Consolidation (DMC), which works well even when the original training data is not available. The idea is to first train a separate model only for the new classes, and then combine the two individual models trained on data of two distinct set of classes (old classes and new classes) via a novel double distillation training objective. The two existing models are consolidated by exploiting publicly available unlabeled auxiliary data. This overcomes the potential difficulties due to the unavailability of original training data. Compared to the state-of-the-art techniques, DMC demonstrates significantly better performance in image classification (CIFAR-100 and CUB-200) and object detection (PASCAL VOC 2007) in the single-headed IL setting.
We propose a framework for learning discrete deterministic planning domains. In this framework, an agent learns the domain by observing the action effects through continuous features that describe the state of the environment after the execution of each action. Besides, the agent learns its perception function, i.e., a probabilistic mapping between state variables and sensor data represented as a vector of continuous random variables called perception variables. We define an algorithm that updates the planning domain and the perception function by (i) introducing new states, either by extending the possible values of state variables, or by weakening their constraints; (ii) adapts the perception function to fit the observed data (iii) adapts the transition function on the basis of the executed actions and the effects observed via the perception function. The framework is able to deal with exogenous events that happen in the environment.
Learning the dynamics of robots from data can help achieve more accurate tracking controllers, or aid their navigation algorithms. However, when the actual dynamics of the robots change due to external conditions, on-line adaptation of their models is required to maintain high fidelity performance. In this work, a framework for on-line learning of robot dynamics is developed to adapt to such changes. The proposed framework employs an incremental support vector regression method to learn the model sequentially from data streams. In combination with the incremental learning, strategies for including and forgetting data are developed to obtain better generalization over the whole state space. The framework is tested in simulation and real experimental scenarios demonstrating its adaptation capabilities to changes in the robot's dynamics.
Conventional approaches to relation extraction usually require a fixed set of pre-defined relations. Such requirement is hard to meet in many real applications, especially when new data and relations are emerging incessantly and it is computationally expensive to store all data and re-train the whole model every time new data and relations come in. We formulate such a challenging problem as lifelong relation extraction and investigate memory-efficient incremental learning methods without catastrophically forgetting knowledge learned from previous tasks. We first investigate a modified version of the stochastic gradient methods with a replay memory, which surprisingly outperforms recent state-of-the-art lifelong learning methods. We further propose to improve this approach to alleviate the forgetting problem by anchoring the sentence embedding space. Specifically, we utilize an explicit alignment model to mitigate the sentence embedding distortion of the learned model when training on new data and new relations. Experiment results on multiple benchmarks show that our proposed method significantly outperforms the state-of-the-art lifelong learning approaches.
In this paper, we treat the image generation task using an autoencoder, a representative latent model. Unlike many studies regularizing the latent variable's distribution by assuming a manually specified prior, we approach the image generation task using an autoencoder by directly estimating the latent distribution. To this end, we introduce 'latent density estimator' which captures latent distribution explicitly and propose its structure. Through experiments, we show that our generative model generates images with the improved visual quality compared to previous autoencoder-based generative models.
We propose an algorithm for incremental learning of classifiers. The proposed method enables an ensemble of classifiers to learn incrementally by accommodating new training data. We use an effective mechanism to overcome the stability-plasticity dilemma. In incremental learning, the general convention is to use only the knowledge acquired in the previous phase but not the previously seen data. We follow this convention by retaining the previously acquired knowledge which is relevant and using it along with the current data. The performance of each classifier is monitored to eliminate the poorly performing classifiers in the subsequent phases. Experimental results show that the proposed approach outperforms the existing incremental learning approaches.
Incremental learning suffers from two challenging problems; forgetting of old knowledge and intransigence on learning new knowledge. Prediction by the model incrementally learned with a subset of the dataset are thus uncertain and the uncertainty accumulates through the tasks by knowledge transfer. To prevent overfitting to the uncertain knowledge, we propose to penalize confident fitting to the uncertain knowledge by the Maximum Entropy Regularizer (MER). Additionally, to reduce class imbalance and induce a self-paced curriculum on new classes, we exclude a few samples from the new classes in every mini-batch, which we call DropOut Sampling (DOS). We further rethink evaluation metrics for forgetting and intransigence in incremental learning by tracking each sample's confusion at the transition of a task since the existing metrics that compute the difference in accuracy are often misleading. We show that the proposed method, named 'MEDIC', outperforms the state-of-the-art incremental learning algorithms in accuracy, forgetting, and intransigence measured by both the existing and the proposed metrics by a large margin in extensive empirical validations on CIFAR100 and a popular subset of ImageNet dataset (TinyImageNet).
We present an algorithm for learning decision trees using stochastic gradient information as the source of supervision. In contrast to previous approaches to gradient-based tree learning, our method operates in the incremental learning setting rather than the batch learning setting, and does not make use of soft splits or require the construction of a new tree for every update. We demonstrate how one can apply these decision trees to different problems by changing only the loss function, using classification, regression, and multi-instance learning as example applications. In the experimental evaluation, our method performs similarly to standard incremental classification trees, outperforms state of the art incremental regression trees, and achieves comparable performance with batch multi-instance learning methods.
Incremental learning (IL) is an important task aimed at increasing the capability of a trained model, in terms of the number of classes recognizable by the model. The key problem in this task is the requirement of storing data (e.g. images) associated with existing classes, while teaching the classifier to learn new classes. However, this is impractical as it increases the memory requirement at every incremental step, which makes it impossible to implement IL algorithms on edge devices with limited memory. Hence, we propose a novel approach, called `Learning without Memorizing (LwM)', to preserve the information about existing (base) classes, without storing any of their data, while making the classifier progressively learn the new classes. In LwM, we present an information preserving penalty: Attention Distillation Loss ($L_{AD}$), and demonstrate that penalizing the changes in classifiers' attention maps helps to retain information of the base classes, as new classes are added. We show that adding $L_{AD}$ to the distillation loss which is an existing information preserving loss consistently outperforms the state-of-the-art performance in the iILSVRC-small and iCIFAR-100 datasets in terms of the overall accuracy of base and incrementally learned classes.
Comprehensive surgical planning require complex patient-specific anatomical models. For instance, functional muskuloskeletal simulations necessitate all relevant structures to be segmented, which could be performed in real-time using deep neural networks given sufficient annotated samples. Such large datasets of multiple structure annotations are costly to procure and are often unavailable in practice. Nevertheless, annotations from different studies and centers can be readily available, or become available in the future in an incremental fashion. We propose a class-incremental segmentation framework for extending a deep network trained for some anatomical structure to yet another structure using a small incremental annotation set. Through distilling knowledge from the current state of the framework, we bypass the need for a full retraining. This is a meta-method to extend any choice of desired deep segmentation network with only a minor addition per structure, which makes it suitable for lifelong class-incremental learning and applicable also for future deep neural network architectures. We evaluated our methods on a public knee dataset of 100 MR volumes. Through varying amount of incremental annotation ratios, we show how our proposed method can retain the previous anatomical structure segmentation performance superior to the conventional finetuning approach. In addition, our framework inherently exploits transferable knowledge from previously trained structures to incremental tasks, demonstrated by superior results compared to non-incremental training. With the presented method, new anatomical structures can be learned without catastrophic forgetting of older structures and without extensive increase of memory and complexity.
Lifelong learning capabilities are crucial for artificial autonomous agents operating on real-world data, which is typically non-stationary and temporally correlated. In this work, we demonstrate that dynamically grown networks outperform static networks in incremental learning scenarios, even when bounded by the same amount of memory in both cases. Learning is unsupervised in our models, a condition that additionally makes training more challenging whilst increasing the realism of the study, since humans are able to learn without dense manual annotation. Our results on artificial neural networks reinforce that structural plasticity constitutes effective prevention against catastrophic forgetting in non-stationary environments, as well as empirically supporting the importance of neurogenesis in the mammalian brain.
In spite of remarkable success of the convolutional neural networks on semantic segmentation, they suffer from catastrophic forgetting: a significant performance drop for the already learned classes when new classes are added on the data, having no annotations for the old classes. We propose an incremental learning methodology, enabling to learn segmenting new classes without hindering dense labeling abilities for the previous classes, although the entire previous data are not accessible. The key points of the proposed approach are adapting the network to learn new as well as old classes on the new training data, and allowing it to remember the previously learned information for the old classes. For adaptation, we keep a frozen copy of the previously trained network, which is used as a memory for the updated network in absence of annotations for the former classes. The updated network minimizes a loss function, which balances the discrepancy between outputs for the previous classes from the memory and updated networks, and the mis-classification rate between outputs for the new classes from the updated network and the new ground-truth. For remembering, we either regularly feed samples from the stored, little fraction of the previous data or use the memory network, depending on whether the new data are collected from completely different geographic areas or from the same city. Our experimental results prove that it is possible to add new classes to the network, while maintaining its performance for the previous classes, despite the whole previous training data are not available.
Modern deep reinforcement learning methods have departed from the incremental learning required for eligibility traces, rendering the implementation of the $\lambda$-return difficult in this context. In particular, off-policy methods that utilize experience replay remain problematic because their random sampling of minibatches is not conducive to the efficient calculation of $\lambda$-returns. Yet replay-based methods are often the most sample efficient, and incorporating $\lambda$-returns into them is a viable way to achieve new state-of-the-art performance. Towards this, we propose the first method to enable practical use of $\lambda$-returns in arbitrary replay-based methods without relying on other forms of decorrelation such as asynchronous gradient updates. By promoting short sequences of past transitions into a small cache within the replay memory, adjacent $\lambda$-returns can be efficiently precomputed by sharing Q-values. Computation is not wasted on experiences that are never sampled, and stored $\lambda$-returns behave as stable temporal-difference (TD) targets that replace the target network. Additionally, our method grants the unique ability to observe TD errors prior to sampling; for the first time, transitions can be prioritized by their true significance rather than by a proxy to it. Furthermore, we propose the novel use of the TD error to dynamically select $\lambda$-values that facilitate faster learning. We show that these innovations can enhance the performance of DQN when playing Atari 2600 games, even under partial observability. While our work specifically focuses on $\lambda$-returns, these ideas are applicable to any multi-step return estimator.
Video object segmentation is an essential task in robot manipulation to facilitate grasping and learning affordances. Incremental learning is important for robotics in unstructured environments, since the total number of objects and their variations can be intractable. Inspired by the children learning process, human robot interaction (HRI) can be utilized to teach robots about the world guided by humans similar to how children learn from a parent or a teacher. A human teacher can show potential objects of interest to the robot, which is able to self adapt to the teaching signal without providing manual segmentation labels. We propose a novel teacher-student learning paradigm to teach robots about their surrounding environment. A two-stream motion and appearance "teacher" network provides pseudo-labels to adapt an appearance "student" network. The student network is able to segment the newly learned objects in other scenes, whether they are static or in motion. We also introduce a carefully designed dataset that serves the proposed HRI setup, denoted as (I)nteractive (V)ideo (O)bject (S)egmentation. Our IVOS dataset contains teaching videos of different objects, and manipulation tasks. Unlike previous datasets, IVOS provides manipulation tasks sequences with segmentation annotation along with the waypoints for the robot trajectories. It also provides segmentation annotation for the different transformations such as translation, scale, planar rotation, and out-of-plane rotation. Our proposed adaptation method outperforms the state-of-the-art on DAVIS and FBMS with 6.8% and 1.2% in F-measure respectively. It improves over the baseline on IVOS dataset with 46.1% and 25.9% in mIoU.
Most of the works on planning and learning, e.g., planning by (model based) reinforcement learning, are based on two main assumptions: (i) the set of states of the planning domain is fixed; (ii) the mapping between the observations from the real word and the states is implicitly assumed or learned offline, and it is not part of the planning domain. Consequently, the focus is on learning the transitions between states. In this paper, we drop such assumptions. We provide a formal framework in which (i) the agent can learn dynamically new states of the planning domain; (ii) the mapping between abstract states and the perception from the real world, represented by continuous variables, is part of the planning domain; (iii) such mapping is learned and updated along the "life" of the agent. We define an algorithm that interleaves planning, acting, and learning, and allows the agent to update the planning domain depending on how much it trusts the model w.r.t. the new experiences learned by executing actions. We define a measure of coherence between the planning domain and the real world as perceived by the agent. We test our approach showing that the agent learns increasingly coherent models, and that the system can scale to deal with models with an order of $10^6$ states.
Many algorithms in workflow scheduling and resource provisioning rely on the performance estimation of tasks to produce a scheduling plan. A profiler that is capable of modeling the execution of tasks and predicting their runtime accurately, therefore, becomes an essential part of any Workflow Management System (WMS). With the emergence of multi-tenant Workflow as a Service (WaaS) platforms that use clouds for deploying scientific workflows, task runtime prediction becomes more challenging because it requires the processing of a significant amount of data in a near real-time scenario while dealing with the performance variability of cloud resources. Hence, relying on methods such as profiling tasks' execution data using basic statistical description (e.g., mean, standard deviation) or batch offline regression techniques to estimate the runtime may not be suitable for such environments. In this paper, we propose an online incremental learning approach to predict the runtime of tasks in scientific workflows in clouds. To improve the performance of the predictions, we harness fine-grained resources monitoring data in the form of time-series records of CPU utilization, memory usage, and I/O activities that are reflecting the unique characteristics of a task's execution. We compare our solution to a state-of-the-art approach that exploits the resources monitoring data based on regression machine learning technique. From our experiments, the proposed strategy improves the performance, in terms of the error, up to 29.89%, compared to the state-of-the-art solutions.
In statistical modelling the biggest threat is concept drift which makes the model gradually showing deteriorating performance over time. There are state of the art methodologies to detect the impact of concept drift, however general strategy considered to overcome the issue in performance is to rebuild or re-calibrate the model periodically as the variable patterns for the model changes significantly due to market change or consumer behavior change etc. Quantitative research is the most widely spread application of data science in Marketing or financial domain where applicability of state of the art reinforcement learning for auto-learning is less explored paradigm. Reinforcement learning is heavily dependent on having a simulated environment which is majorly available for gaming or online systems, to learn from the live feedback. However, there are some research happened on the area of online advertisement, pricing etc where due to the nature of the online learning environment scope of reinforcement learning is explored. Our proposed solution is a reinforcement learning based, true self-learning algorithm which can adapt to the data change or concept drift and auto learn and self-calibrate for the new patterns of the data solving the problem of concept drift. Keywords - Reinforcement learning, Genetic Algorithm, Q-learning, Classification modelling, CMA-ES, NES, Multi objective optimization, Concept drift, Population stability index, Incremental learning, F1-measure, Predictive Modelling, Self-learning, MCTS, AlphaGo, AlphaZero
Deep learning-based methods have reached state of the art performances, relying on large quantity of available data and computational power. Such methods still remain highly inappropriate when facing a major open machine learning problem, which consists of learning incrementally new classes and examples over time. Combining the outstanding performances of Deep Neural Networks (DNNs) with the flexibility of incremental learning techniques is a promising venue of research. In this contribution, we introduce Transfer Incremental Learning using Data Augmentation (TILDA). TILDA is based on pre-trained DNNs as feature extractor, robust selection of feature vectors in subspaces using a nearest-class-mean based technique, majority votes and data augmentation at both the training and the prediction stages. Experiments on challenging vision datasets demonstrate the ability of the proposed method for low complexity incremental learning, while achieving significantly better accuracy than existing incremental counterparts.
The great success that deep models have achieved in the past is mainly owed to large amounts of labeled training data. However, the acquisition of labeled data for new tasks aside from existing benchmarks is both challenging and costly. Active learning can make the process of labeling new data more efficient by selecting unlabeled samples which, when labeled, are expected to improve the model the most. In this paper, we combine a novel method of active learning for object detection with an incremental learning scheme to enable continuous exploration of new unlabeled datasets. We propose a set of uncertainty-based active learning metrics suitable for most object detectors. Furthermore, we present an approach to leverage class imbalances during sample selection. All methods are evaluated systematically in a continuous exploration context on the PASCAL VOC 2012 dataset.
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 environments. An automated construction of a denoising autoeconder, namely deep evolving denoising autoencoder (DEVDAN), is proposed in this paper. DEVDAN features an open structure both in the generative phase and in the discriminative phase where input features can be automatically added and discarded on the fly. A network significance (NS) method is formulated in this paper and is derived from the bias-variance concept. This method is capable of estimating the statistical contribution of the network structure and its hidden units which precursors an ideal state to add or prune input features. Furthermore, DEVDAN is free of the problem- specific threshold and works fully in the single-pass learning fashion. The efficacy of DEVDAN is numerically validated using nine non-stationary data stream problems simulated under the prequential test-then-train protocol where DEVDAN is capable of delivering an improvement of classification accuracy to recently published online learning works while having flexibility in the automatic extraction of robust input features and in adapting to rapidly changing environments.