Julio Hurtado

Albin Soutif--Cormerais, Antonio Carta, Andrea Cossu et al.

Online continual learning aims to get closer to a live learning experience by learning directly on a stream of data with temporally shifting distribution and by storing a minimum amount of data from that stream. In this empirical evaluation, we evaluate various methods from the literature that tackle online continual learning. More specifically, we focus on the class-incremental setting in the context of image classification, where the learner must learn new classes incrementally from a stream of data. We compare these methods on the Split-CIFAR100 and Split-TinyImagenet benchmarks, and measure their average accuracy, forgetting, stability, and quality of the representations, to evaluate various aspects of the algorithm at the end but also during the whole training period. We find that most methods suffer from stability and underfitting issues. However, the learned representations are comparable to i.i.d. training under the same computational budget. No clear winner emerges from the results and basic experience replay, when properly tuned and implemented, is a very strong baseline. We release our modular and extensible codebase at https://github.com/AlbinSou/ocl_survey based on the avalanche framework to reproduce our results and encourage future research.

Hamed Hemati, Andrea Cossu, Antonio Carta et al. · berkeley

Real-world data streams naturally include the repetition of previous concepts. From a Continual Learning (CL) perspective, repetition is a property of the environment and, unlike replay, cannot be controlled by the agent. Nowadays, the Class-Incremental (CI) scenario represents the leading test-bed for assessing and comparing CL strategies. This scenario type is very easy to use, but it never allows revisiting previously seen classes, thus completely neglecting the role of repetition. We focus on the family of Class-Incremental with Repetition (CIR) scenario, where repetition is embedded in the definition of the stream. We propose two stochastic stream generators that produce a wide range of CIR streams starting from a single dataset and a few interpretable control parameters. We conduct the first comprehensive evaluation of repetition in CL by studying the behavior of existing CL strategies under different CIR streams. We then present a novel replay strategy that exploits repetition and counteracts the natural imbalance present in the stream. On both CIFAR100 and TinyImageNet, our strategy outperforms other replay approaches, which are not designed for environments with repetition.

Andrés Villa, Juan León Alcázar, Motasem Alfarra et al.

Modern machine learning pipelines are limited due to data availability, storage quotas, privacy regulations, and expensive annotation processes. These constraints make it difficult or impossible to train and update large-scale models on such dynamic annotated sets. Continual learning directly approaches this problem, with the ultimate goal of devising methods where a deep neural network effectively learns relevant patterns for new (unseen) classes, without significantly altering its performance on previously learned ones. In this paper, we address the problem of continual learning for video data. We introduce PIVOT, a novel method that leverages extensive knowledge in pre-trained models from the image domain, thereby reducing the number of trainable parameters and the associated forgetting. Unlike previous methods, ours is the first approach that effectively uses prompting mechanisms for continual learning without any in-domain pre-training. Our experiments show that PIVOT improves state-of-the-art methods by a significant 27% on the 20-task ActivityNet setup.

Julio Hurtado, Dario Salvati, Rudy Semola et al.

Deep learning techniques have become one of the main propellers for solving engineering problems effectively and efficiently. For instance, Predictive Maintenance methods have been used to improve predictions of when maintenance is needed on different machines and operative contexts. However, deep learning methods are not without limitations, as these models are normally trained on a fixed distribution that only reflects the current state of the problem. Due to internal or external factors, the state of the problem can change, and the performance decreases due to the lack of generalization and adaptation. Contrary to this stationary training set, real-world applications change their environments constantly, creating the need to constantly adapt the model to evolving scenarios. To aid in this endeavor, Continual Learning methods propose ways to constantly adapt prediction models and incorporate new knowledge after deployment. Despite the advantages of these techniques, there are still challenges to applying them to real-world problems. In this work, we present a brief introduction to predictive maintenance, non-stationary environments, and continual learning, together with an extensive review of the current state of applying continual learning in real-world applications and specifically in predictive maintenance. We then discuss the current challenges of both predictive maintenance and continual learning, proposing future directions at the intersection of both areas. Finally, we propose a novel way to create benchmarks that favor the application of continuous learning methods in more realistic environments, giving specific examples of predictive maintenance.

Julio Hurtado, Alain Raymond-Saez, Vladimir Araujo et al.

Catastrophic forgetting, the phenomenon of forgetting previously learned tasks when learning a new one, is a major hurdle in developing continual learning algorithms. A popular method to alleviate forgetting is to use a memory buffer, which stores a subset of previously learned task examples for use during training on new tasks. The de facto method of filling memory is by randomly selecting previous examples. However, this process could introduce outliers or noisy samples that could hurt the generalization of the model. This paper introduces Memory Outlier Elimination (MOE), a method for identifying and eliminating outliers in the memory buffer by choosing samples from label-homogeneous subpopulations. We show that a space with a high homogeneity is related to a feature space that is more representative of the class distribution. In practice, MOE removes a sample if it is surrounded by samples from different labels. We demonstrate the effectiveness of MOE on CIFAR-10, CIFAR-100, and CORe50, outperforming previous well-known memory population methods.

Vladimir Araujo, Julio Hurtado, Alvaro Soto et al.

The ability to continuously learn remains elusive for deep learning models. Unlike humans, models cannot accumulate knowledge in their weights when learning new tasks, mainly due to an excess of plasticity and the low incentive to reuse weights when training a new task. To address the stability-plasticity dilemma in neural networks, we propose a novel method called Entropy-based Stability-Plasticity (ESP). Our approach can decide dynamically how much each model layer should be modified via a plasticity factor. We incorporate branch layers and an entropy-based criterion into the model to find such factor. Our experiments in the domains of natural language and vision show the effectiveness of our approach in leveraging prior knowledge by reducing interference. Also, in some cases, it is possible to freeze layers during training leading to speed up in training.

Vladimir Araujo, Helena Balabin, Julio Hurtado et al.

Lifelong language learning seeks to have models continuously learn multiple tasks in a sequential order without suffering from catastrophic forgetting. State-of-the-art approaches rely on sparse experience replay as the primary approach to prevent forgetting. Experience replay usually adopts sampling methods for the memory population; however, the effect of the chosen sampling strategy on model performance has not yet been studied. In this paper, we investigate how relevant the selective memory population is in the lifelong learning process of text classification and question-answering tasks. We found that methods that randomly store a uniform number of samples from the entire data stream lead to high performances, especially for low memory size, which is consistent with computer vision studies.

Felipe del Rio, Julio Hurtado, Cristian Buc et al.

One of the objectives of Continual Learning is to learn new concepts continually over a stream of experiences and at the same time avoid catastrophic forgetting. To mitigate complete knowledge overwriting, memory-based methods store a percentage of previous data distributions to be used during training. Although these methods produce good results, few studies have tested their out-of-distribution generalization properties, as well as whether these methods overfit the replay memory. In this work, we show that although these methods can help in traditional in-distribution generalization, they can strongly impair out-of-distribution generalization by learning spurious features and correlations. Using a controlled environment, the Synbol benchmark generator (Lacoste et al., 2020), we demonstrate that this lack of out-of-distribution generalization mainly occurs in the linear classifier.

Eric Nuertey Coleman, Julio Hurtado, Vincenzo Lomonaco

Large language models (LLMs) have had a huge impact on society due to their impressive capabilities and vast knowledge of the world. Various applications and tools have been created that allow users to interact with these models in a black-box scenario. However, one limitation of this scenario is that users cannot modify the internal knowledge of the model, and the only way to add or modify internal knowledge is by explicitly mentioning it to the model during the current interaction. This learning process is called in-context training, and it refers to training that is confined to the user's current session or context. In-context learning has significant applications, but also has limitations that are seldom studied. In this paper, we present a study that shows how the model can suffer from interference between information that continually flows in the context, causing it to forget previously learned knowledge, which can reduce the model's performance. Along with showing the problem, we propose an evaluation benchmark based on the bAbI dataset.

Clea Rebillard, Julio Hurtado, Andrii Krutsylo et al.

Learning continually from a stream of non-i.i.d. data is an open challenge in deep learning, even more so when working in resource-constrained environments such as embedded devices. Visual models that are continually updated through supervised learning are often prone to overfitting, catastrophic forgetting, and biased representations. On the other hand, large language models contain knowledge about multiple concepts and their relations, which can foster a more robust, informed and coherent learning process. This work proposes Continual Visual Mapping (CVM), an approach that continually ground vision representations to a knowledge space extracted from a fixed Language model. Specifically, CVM continually trains a small and efficient visual model to map its representations into a conceptual space established by a fixed Large Language Model. Due to their smaller nature, CVM can be used when directly adapting large visual pre-trained models is unfeasible due to computational or data constraints. CVM overcome state-of-the-art continual learning methods on five benchmarks and offers a promising avenue for addressing generalization capabilities in continual learning, even in computationally constrained devices.

Alain Raymond-Saez, Julio Hurtado, Alvaro Soto

Curriculum Learning is a powerful training method that allows for faster and better training in some settings. This method, however, requires having a notion of which examples are difficult and which are easy, which is not always trivial to provide. A recent metric called C-Score acts as a proxy for example difficulty by relating it to learning consistency. Unfortunately, this method is quite compute intensive which limits its applicability for alternative datasets. In this work, we train models through different methods to predict C-Score for CIFAR-100 and CIFAR-10. We find, however, that these models generalize poorly both within the same distribution as well as out of distribution. This suggests that C-Score is not defined by the individual characteristics of each sample but rather by other factors. We hypothesize that a sample's relation to its neighbours, in particular, how many of them share the same labels, can help in explaining C-Scores. We plan to explore this in future work.

Ali Hummos, Felipe del Río, Brabeeba Mien Wang et al.

Humans and many animals show remarkably adaptive behavior and can respond differently to the same input depending on their internal goals. The brain not only represents the intermediate abstractions needed to perform a computation but also actively maintains a representation of the computation itself (task abstraction). Such separation of the computation and its abstraction is associated with faster learning, flexible decision-making, and broad generalization capacity. We investigate if such benefits might extend to neural networks trained with task abstractions. For such benefits to emerge, one needs a task inference mechanism that possesses two crucial abilities: First, the ability to infer abstract task representations when no longer explicitly provided (task inference), and second, manipulate task representations to adapt to novel problems (task recomposition). To tackle this, we cast task inference as an optimization problem from a variational inference perspective and ground our approach in an expectation-maximization framework. We show that gradients backpropagated through a neural network to a task representation layer are an efficient heuristic to infer current task demands, a process we refer to as gradient-based inference (GBI). Further iterative optimization of the task representation layer allows for recomposing abstractions to adapt to novel situations. Using a toy example, a novel image classifier, and a language model, we demonstrate that GBI provides higher learning efficiency and generalization to novel tasks and limits forgetting. Moreover, we show that GBI has unique advantages such as preserving information for uncertainty estimation and detecting out-of-distribution samples.

Fabrice Normandin, Florian Golemo, Oleksiy Ostapenko et al.

The field of Continual Learning (CL) seeks to develop algorithms that accumulate knowledge and skills over time through interaction with non-stationary environments. In practice, a plethora of evaluation procedures (settings) and algorithmic solutions (methods) exist, each with their own potentially disjoint set of assumptions. This variety makes measuring progress in CL difficult. We propose a taxonomy of settings, where each setting is described as a set of assumptions. A tree-shaped hierarchy emerges from this view, where more general settings become the parents of those with more restrictive assumptions. This makes it possible to use inheritance to share and reuse research, as developing a method for a given setting also makes it directly applicable onto any of its children. We instantiate this idea as a publicly available software framework called Sequoia, which features a wide variety of settings from both the Continual Supervised Learning (CSL) and Continual Reinforcement Learning (CRL) domains. Sequoia also includes a growing suite of methods which are easy to extend and customize, in addition to more specialized methods from external libraries. We hope that this new paradigm and its first implementation can help unify and accelerate research in CL. You can help us grow the tree by visiting www.github.com/lebrice/Sequoia.

Rudy Semola, Julio Hurtado, Vincenzo Lomonaco et al.

Hyperparameter selection in continual learning scenarios is a challenging and underexplored aspect, especially in practical non-stationary environments. Traditional approaches, such as grid searches with held-out validation data from all tasks, are unrealistic for building accurate lifelong learning systems. This paper aims to explore the role of hyperparameter selection in continual learning and the necessity of continually and automatically tuning them according to the complexity of the task at hand. Hence, we propose leveraging the nature of sequence task learning to improve Hyperparameter Optimization efficiency. By using the functional analysis of variance-based techniques, we identify the most crucial hyperparameters that have an impact on performance. We demonstrate empirically that this approach, agnostic to continual scenarios and strategies, allows us to speed up hyperparameters optimization continually across tasks and exhibit robustness even in the face of varying sequential task orders. We believe that our findings can contribute to the advancement of continual learning methodologies towards more efficient, robust and adaptable models for real-world applications.

Eric Nuertey Coleman, Luigi Quarantiello, Ziyue Liu et al.

The emergence of large pre-trained networks has revolutionized the AI field, unlocking new possibilities and achieving unprecedented performance. However, these models inherit a fundamental limitation from traditional Machine Learning approaches: their strong dependence on the \textit{i.i.d.} assumption hinders their adaptability to dynamic learning scenarios. We believe the next breakthrough in AI lies in enabling efficient adaptation to evolving environments -- such as the real world -- where new data and tasks arrive sequentially. This challenge defines the field of Continual Learning (CL), a Machine Learning paradigm focused on developing lifelong learning neural models. One alternative to efficiently adapt these large-scale models is known Parameter-Efficient Fine-Tuning (PEFT). These methods tackle the issue of adapting the model to a particular data or scenario by performing small and efficient modifications, achieving similar performance to full fine-tuning. However, these techniques still lack the ability to adjust the model to multiple tasks continually, as they suffer from the issue of Catastrophic Forgetting. In this survey, we first provide an overview of CL algorithms and PEFT methods before reviewing the state-of-the-art on Parameter-Efficient Continual Fine-Tuning (PECFT). We examine various approaches, discuss evaluation metrics, and explore potential future research directions. Our goal is to highlight the synergy between CL and Parameter-Efficient Fine-Tuning, guide researchers in this field, and pave the way for novel future research directions.

Eric Nuertey Coleman, Luigi Quarantiello, Samrat Mukherjee et al.

Continual learning is an essential capability of human cognition, yet it poses significant challenges for current deep learning models. The primary issue is that new knowledge can interfere with previously learned information, causing the model to forget earlier knowledge in favor of the new, a phenomenon known as catastrophic forgetting. Although large pre-trained models can partially mitigate forgetting by leveraging their existing knowledge and over-parameterization, they often struggle when confronted with novel data distributions. Parameter-Efficient Fine-Tuning (PEFT) methods, such as LoRA, enable efficient adaptation to new knowledge. However, they still face challenges in scaling to dynamic learning scenarios and long sequences of tasks, as maintaining one adapter per task introduces complexity and increases the potential for interference. In this paper, we introduce Hierarchical Adapters Merging (HAM), a novel framework that dynamically combines adapters from different tasks during training. This approach enables HAM to scale effectively, allowing it to manage more tasks than competing baselines with improved efficiency. To achieve this, HAM maintains a fixed set of groups that hierarchically consolidate new adapters. For each task, HAM trains a low-rank adapter along with an importance scalar, then dynamically groups tasks based on adapter similarity. Within each group, adapters are pruned, scaled and merge, facilitating transfer learning between related tasks. Extensive experiments on three vision benchmarks show that HAM significantly outperforms state-of-the-art methods, particularly as the number of tasks increases.

Felipe del Rio, Alain Raymond-Saez, Daniel Florea et al.

Deep neural networks (DNNs) struggle at systematic generalization (SG). Several studies have evaluated the possibility to promote SG through the proposal of novel architectures, loss functions or training methodologies. Few studies, however, have focused on the role of training data properties in promoting SG. In this work, we investigate the impact of certain data distributional properties, as inductive biases for the SG ability of a multi-modal language model. To this end, we study three different properties. First, data diversity, instantiated as an increase in the possible values a latent property in the training distribution may take. Second, burstiness, where we probabilistically restrict the number of possible values of latent factors on particular inputs during training. Third, latent intervention, where a particular latent factor is altered randomly during training. We find that all three factors significantly enhance SG, with diversity contributing an 89% absolute increase in accuracy in the most affected property. Through a series of experiments, we test various hypotheses to understand why these properties promote SG. Finally, we find that Normalized Mutual Information (NMI) between latent attributes in the training distribution is strongly predictive of out-of-distribution generalization. We find that a mechanism by which lower NMI induces SG is in the geometry of representations. In particular, we find that NMI induces more parallelism in neural representations (i.e., input features coded in parallel neural vectors) of the model, a property related to the capacity of reasoning by analogy.

Julio Hurtado, Haoran Ni, Duygu Sap et al. · berkeley, oxford

The fashion industry has been identified as a major contributor to waste and emissions, leading to an increased interest in promoting the second-hand market. Machine learning methods play an important role in facilitating the creation and expansion of second-hand marketplaces by enabling the large-scale valuation of used garments. We contribute to this line of work by addressing the scalability of second-hand image retrieval from databases. By introducing a selective representation framework, we can shrink databases to 10% of their original size without sacrificing retrieval accuracy. We first explore clustering and coreset selection methods to identify representative samples that capture the key features of each garment and its internal variability. Then, we introduce an efficient outlier removal method, based on a neighbour-homogeneity consistency score measure, that filters out uncharacteristic samples prior to selection. We evaluate our approach on three public datasets: DeepFashion Attribute, DeepFashion Con2Shop, and DeepFashion2. The results demonstrate a clear performance-efficiency trade-off by strategically pruning and selecting representative vectors of images. The retrieval system maintains near-optimal accuracy, while greatly reducing computational costs by reducing the images added to the vector database. Furthermore, applying our outlier removal method to clustering techniques yields even higher retrieval performance by removing non-discriminative samples before the selection.

Julio Hurtado, Alain Raymond-Saez, Alvaro Soto

When learning tasks over time, artificial neural networks suffer from a problem known as Catastrophic Forgetting (CF). This happens when the weights of a network are overwritten during the training of a new task causing forgetting of old information. To address this issue, we propose MetA Reusable Knowledge or MARK, a new method that fosters weight reusability instead of overwriting when learning a new task. Specifically, MARK keeps a set of shared weights among tasks. We envision these shared weights as a common Knowledge Base (KB) that is not only used to learn new tasks, but also enriched with new knowledge as the model learns new tasks. Key components behind MARK are two-fold. On the one hand, a metalearning approach provides the key mechanism to incrementally enrich the KB with new knowledge and to foster weight reusability among tasks. On the other hand, a set of trainable masks provides the key mechanism to selectively choose from the KB relevant weights to solve each task. By using MARK, we achieve state of the art results in several popular benchmarks, surpassing the best performing methods in terms of average accuracy by over 10% on the 20-Split-MiniImageNet dataset, while achieving almost zero forgetfulness using 55% of the number of parameters. Furthermore, an ablation study provides evidence that, indeed, MARK is learning reusable knowledge that is selectively used by each task.