Davide Maltoni

Matteo Ferrara, Annalisa Franco, Davide Maltoni et al.

In security systems the risk assessment in the sense of common criteria testing is a very relevant topic; this requires quantifying the attack potential in terms of the expertise of the attacker, his knowledge about the target and access to equipment. Contrary to those attacks, the recently revealed morphing attacks against Face Recognition Systems (FRSs) can not be assessed by any of the above criteria. But not all morphing techniques pose the same risk for an operational face recognition system. This paper introduces with the Morphing Attack Potential (MAP) a consistent methodology, that can quantify the risk, which a certain morphing attack creates.

Gabriele Graffieti, Davide Maltoni, Lorenzo Pellegrini et al.

Learning continually is a key aspect of intelligence and a necessary ability to solve many real-life problems. One of the most effective strategies to control catastrophic forgetting, the Achilles' heel of continual learning, is storing part of the old data and replaying them interleaved with new experiences (also known as the replay approach). Generative replay, which is using generative models to provide replay patterns on demand, is particularly intriguing, however, it was shown to be effective mainly under simplified assumptions, such as simple scenarios and low-dimensional data. In this paper, we show that, while the generated data are usually not able to improve the classification accuracy for the old classes, they can be effective as negative examples (or antagonists) to better learn the new classes, especially when the learning experiences are small and contain examples of just one or few classes. The proposed approach is validated on complex class-incremental and data-incremental continual learning scenarios (CORe50 and ImageNet-1000) composed of high-dimensional data and a large number of training experiences: a setup where existing generative replay approaches usually fail.

Davide Maltoni, Matteo Ferrara

A better understanding of the emergent computation and problem-solving capabilities of recent large language models is of paramount importance to further improve them and broaden their applicability. This work investigates how a language model, trained to predict the next token, can perform arithmetic computations generalizing beyond training data. Binary addition and multiplication constitute a good testbed for this purpose, since they require a very small vocabulary and exhibit relevant input/output discontinuities making smooth input interpolation ineffective for novel data. We successfully trained a light language model to learn these tasks and ran a number of experiments to investigate the extrapolation capabilities and internal information processing. Our findings support the hypothesis that the language model works as an Encoding-Regression-Decoding machine where the computation takes place in the value space once the input token representation is mapped to an appropriate internal representation.

Lorenzo Pellegrini, Guido Borghi, Annalisa Franco et al.

Scenarios in which restrictions in data transfer and storage limit the possibility to compose a single dataset -- also exploiting different data sources -- to perform a batch-based training procedure, make the development of robust models particularly challenging. We hypothesize that the recent Continual Learning (CL) paradigm may represent an effective solution to enable incremental training, even through multiple sites. Indeed, a basic assumption of CL is that once a model has been trained, old data can no longer be used in successive training iterations and in principle can be deleted. Therefore, in this paper, we investigate the performance of different Continual Learning methods in this scenario, simulating a learning model that is updated every time a new chunk of data, even of variable size, is available. Experimental results reveal that a particular CL method, namely Learning without Forgetting (LwF), is one of the best-performing algorithms. Then, we investigate its usage and parametrization in Morphing Attack Detection and Object Classification tasks, specifically with respect to the amount of new training data that became available.

Lorenzo Vorabbi, Davide Maltoni, Stefano Santi

Existing Continual Learning (CL) solutions only partially address the constraints on power, memory and computation of the deep learning models when deployed on low-power embedded CPUs. In this paper, we propose a CL solution that embraces the recent advancements in CL field and the efficiency of the Binary Neural Networks (BNN), that use 1-bit for weights and activations to efficiently execute deep learning models. We propose a hybrid quantization of CWR* (an effective CL approach) that considers differently forward and backward pass in order to retain more precision during gradient update step and at the same time minimizing the latency overhead. The choice of a binary network as backbone is essential to meet the constraints of low power devices and, to the best of authors' knowledge, this is the first attempt to prove on-device learning with BNN. The experimental validation carried out confirms the validity and the suitability of the proposed method.

Guido Borghi, Gabriele Graffieti, Davide Maltoni

In real-world contexts, sometimes data are available in form of Natural Data Streams, i.e. data characterized by a streaming nature, unbalanced distribution, data drift over a long time frame and strong correlation of samples in short time ranges. Moreover, a clear separation between the traditional training and deployment phases is usually lacking. This data organization and fruition represents an interesting and challenging scenario for both traditional Machine and Deep Learning algorithms and incremental learning agents, i.e. agents that have the ability to incrementally improve their knowledge through the past experience. In this paper, we investigate the classification performance of a variety of algorithms that belong to various research field, i.e. Continual, Streaming and Online Learning, that receives as training input Natural Data Streams. The experimental validation is carried out on three different datasets, expressly organized to replicate this challenging setting.

Vincenzo Lomonaco, Lorenzo Pellegrini, Gabriele Graffieti et al.

In recent years we have witnessed a renewed interest in machine learning methodologies, especially for deep representation learning, that could overcome basic i.i.d. assumptions and tackle non-stationary environments subject to various distributional shifts or sample selection biases. Within this context, several computational approaches based on architectural priors, regularizers and replay policies have been proposed with different degrees of success depending on the specific scenario in which they were developed and assessed. However, designing comprehensive hybrid solutions that can flexibly and generally be applied with tunable efficiency-effectiveness trade-offs still seems a distant goal. In this paper, we propose "Architect, Regularize and Replay" (ARR), an hybrid generalization of the renowned AR1 algorithm and its variants, that can achieve state-of-the-art results in classic scenarios (e.g. class-incremental learning) but also generalize to arbitrary data streams generated from real-world datasets such as CIFAR-100, CORe50 and ImageNet-1000.

Nicolò Di Domenico, Annalisa Franco, Matteo Ferrara et al.

Face morphing attacks are widely recognized as one of the most challenging threats to face recognition systems used in electronic identity documents. These attacks exploit a critical vulnerability in passport enrollment procedures adopted by many countries, where the facial image is often acquired without a supervised live capture process. In this paper, we propose a novel face morphing technique based on Arc2Face, an identity-conditioned face foundation model capable of synthesizing photorealistic facial images from compact identity representations. We demonstrate the effectiveness of the proposed approach by comparing the morphing attack potential metric on two large-scale sequestered face morphing attack detection datasets against several state-of-the-art morphing methods, as well as on two novel morphed face datasets derived from FEI and ONOT. Experimental results show that the proposed deep learning-based approach achieves a morphing attack potential comparable to that of landmark-based techniques, which have traditionally been regarded as the most challenging. These findings confirm the ability of the proposed method to effectively preserve and manage identity information during the morph generation process.

Lorenzo Pellegrini, Serafino Pandolfini, Davide Maltoni et al.

The effectiveness of deepfake detection methods often depends less on their core design and more on implementation details such as data preprocessing, augmentation strategies, and optimization techniques. These factors make it difficult to fairly compare detectors and to understand which factors truly contribute to their performance. To address this, we systematically investigate how different design choices influence the accuracy and generalization capabilities of deepfake detection models, focusing on aspects related to training, inference, and incremental updates. By isolating the impact of individual factors, we aim to establish robust, architecture-agnostic best practices for the design and development of future deepfake detection systems. Our experiments identify a set of design choices that consistently improve deepfake detection and enable state-of-the-art performance on the AI-GenBench benchmark.

Serafino Pandolfini, Lorenzo Pellegrini, Matteo Ferrara et al.

The rapid progress of generative AI has enabled highly realistic image manipulations, including inpainting and region-level editing. These approaches preserve most of the original visual context and are increasingly exploited in cybersecurity-relevant threat scenarios. While numerous detectors have been proposed for identifying fully synthetic images, their ability to generalize to localized manipulations remains insufficiently characterized. This work presents a systematic evaluation of state-of-the-art detectors, originally trained for the deepfake detection on fully synthetic images, when applied to a distinct challenge: localized inpainting detection. The study leverages multiple datasets spanning diverse generators, mask sizes, and inpainting techniques. Our experiments show that models trained on a large set of generators exhibit partial transferability to inpainting-based edits and can reliably detect medium- and large-area manipulations or regeneration-style inpainting, outperforming many existing ad hoc detection approaches.

Matteo Scucchia, Davide Maltoni

Recognizing already explored places (a.k.a. place recognition) is a fundamental task in Simultaneous Localization and Mapping (SLAM) to enable robot relocalization and loop closure detection. In topological SLAM the recognition takes place by comparing a signature (or feature vector) associated to the current node with the signatures of the nodes in the known map. However, as the number of nodes increases, matching the current node signature against all the existing ones becomes inefficient and thwarts real-time navigation. In this paper we propose a novel approach to pre-select a subset of map nodes for place recognition. The map nodes are clustered during exploration and each cluster is associated with a region. The region labels become the prediction targets of a deep neural network and, during navigation, only the nodes associated with the regions predicted with high probability are considered for matching. While the proposed technique can be integrated in different SLAM approaches, in this work we describe an effective integration with RTAB-Map (a popular framework for real-time topological SLAM) which allowed us to design and run several experiments to demonstrate its effectiveness. All the code and material from the experiments will be available online at https://github.com/MI-BioLab/region-learner.

Vincenzo Lomonaco, Lorenzo Pellegrini, Andrea Cossu et al.

Learning continually from non-stationary data streams is a long-standing goal and a challenging problem in machine learning. Recently, we have witnessed a renewed and fast-growing interest in continual learning, especially within the deep learning community. However, algorithmic solutions are often difficult to re-implement, evaluate and port across different settings, where even results on standard benchmarks are hard to reproduce. In this work, we propose Avalanche, an open-source end-to-end library for continual learning research based on PyTorch. Avalanche is designed to provide a shared and collaborative codebase for fast prototyping, training, and reproducible evaluation of continual learning algorithms.

Hatef Otroshi Shahreza, Christophe Ecabert, Anjith George et al.

Large-scale face recognition datasets are collected by crawling the Internet and without individuals' consent, raising legal, ethical, and privacy concerns. With the recent advances in generative models, recently several works proposed generating synthetic face recognition datasets to mitigate concerns in web-crawled face recognition datasets. This paper presents the summary of the Synthetic Data for Face Recognition (SDFR) Competition held in conjunction with the 18th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2024) and established to investigate the use of synthetic data for training face recognition models. The SDFR competition was split into two tasks, allowing participants to train face recognition systems using new synthetic datasets and/or existing ones. In the first task, the face recognition backbone was fixed and the dataset size was limited, while the second task provided almost complete freedom on the model backbone, the dataset, and the training pipeline. The submitted models were trained on existing and also new synthetic datasets and used clever methods to improve training with synthetic data. The submissions were evaluated and ranked on a diverse set of seven benchmarking datasets. The paper gives an overview of the submitted face recognition models and reports achieved performance compared to baseline models trained on real and synthetic datasets. Furthermore, the evaluation of submissions is extended to bias assessment across different demography groups. Lastly, an outlook on the current state of the research in training face recognition models using synthetic data is presented, and existing problems as well as potential future directions are also discussed.

Nicolò Di Domenico, Guido Borghi, Annalisa Franco et al.

Nowadays, state-of-the-art AI-based generative models represent a viable solution to overcome privacy issues and biases in the collection of datasets containing personal information, such as faces. Following this intuition, in this paper we introduce ONOT, a synthetic dataset specifically focused on the generation of high-quality faces in adherence to the requirements of the ISO/IEC 39794-5 standards that, following the guidelines of the International Civil Aviation Organization (ICAO), defines the interchange formats of face images in electronic Machine-Readable Travel Documents (eMRTD). The strictly controlled and varied mugshot images included in ONOT are useful in research fields related to the analysis of face images in eMRTD, such as Morphing Attack Detection and Face Quality Assessment. The dataset is publicly released, in combination with the generation procedure details in order to improve the reproducibility and enable future extensions.

Matteo Scucchia, Matteo Ferrara, Davide Maltoni

In computer vision, the development of robust algorithms capable of generalizing effectively in real-world scenarios more and more often requires large-scale datasets collected under diverse environmental conditions. However, acquiring such datasets is time-consuming, costly, and sometimes unfeasible. To address these limitations, the use of synthetic data has gained attention as a viable alternative, allowing researchers to generate vast amounts of data while simulating various environmental contexts in a controlled setting. In this study, we investigate the use of synthetic data in robotics and navigation, specifically focusing on Simultaneous Localization and Mapping (SLAM) and Visual Place Recognition (VPR). In particular, we introduce a synthetic dataset created using the virtual environment of the video game Grand Theft Auto V (GTA V), along with an algorithm designed to generate a VPR dataset, without human supervision. Through a series of experiments centered on SLAM and VPR, we demonstrate that synthetic data derived from GTA V are qualitatively comparable to real-world data. Furthermore, these synthetic data can complement or even substitute real-world data in these applications. This study sets the stage for the creation of large-scale synthetic datasets, offering a cost-effective and scalable solution for future research and development.

Nicolò Di Domenico, Guido Borghi, Annalisa Franco et al.

The advent of morphing attacks has posed significant security concerns for automated Face Recognition systems, raising the pressing need for robust and effective Morphing Attack Detection (MAD) methods able to effectively address this issue. In this paper, we focus on Differential MAD (D-MAD), where a trusted live capture, usually representing the criminal, is compared with the document image to classify it as morphed or bona fide. We show these approaches based on identity features are effective when the morphed image and the live one are sufficiently diverse; unfortunately, the effectiveness is significantly reduced when the same approaches are applied to look-alike subjects or in all those cases when the similarity between the two compared images is high (e.g. comparison between the morphed image and the accomplice). Therefore, in this paper, we propose ACIdA, a modular D-MAD system, consisting of a module for the attempt type classification, and two modules for the identity and artifacts analysis on input images. Successfully addressing this task would allow broadening the D-MAD applications including, for instance, the document enrollment stage, which currently relies entirely on human evaluation, thus limiting the possibility of releasing ID documents with manipulated images, as well as the automated gates to detect both accomplices and criminals. An extensive cross-dataset experimental evaluation conducted on the introduced scenario shows that ACIdA achieves state-of-the-art results, outperforming literature competitors, while maintaining good performance in traditional D-MAD benchmarks.

Guido Borghi, Annalisa Franco, Nicolò Di Domenico et al.

In response to the rising threat of the face morphing attack, this paper introduces and explores the potential of Video-based Morphing Attack Detection (V-MAD) systems in real-world operational scenarios. While current morphing attack detection methods primarily focus on a single or a pair of images, V-MAD is based on video sequences, exploiting the video streams often acquired by face verification tools available, for instance, at airport gates. Through this study, we show for the first time the advantages that the availability of multiple probe frames can bring to the morphing attack detection task, especially in scenarios where the quality of probe images is varied and might be affected, for instance, by pose or illumination variations. Experimental results on a real operational database demonstrate that video sequences represent valuable information for increasing the robustness and performance of morphing attack detection systems.

Lorenzo Pellegrini, Davide Cozzolino, Serafino Pandolfini et al.

The rapid advancement of generative AI has revolutionized image creation, enabling high-quality synthesis from text prompts while raising critical challenges for media authenticity. We present Ai-GenBench, a novel benchmark designed to address the urgent need for robust detection of AI-generated images in real-world scenarios. Unlike existing solutions that evaluate models on static datasets, Ai-GenBench introduces a temporal evaluation framework where detection methods are incrementally trained on synthetic images, historically ordered by their generative models, to test their ability to generalize to new generative models, such as the transition from GANs to diffusion models. Our benchmark focuses on high-quality, diverse visual content and overcomes key limitations of current approaches, including arbitrary dataset splits, unfair comparisons, and excessive computational demands. Ai-GenBench provides a comprehensive dataset, a standardized evaluation protocol, and accessible tools for both researchers and non-experts (e.g., journalists, fact-checkers), ensuring reproducibility while maintaining practical training requirements. By establishing clear evaluation rules and controlled augmentation strategies, Ai-GenBench enables meaningful comparison of detection methods and scalable solutions. Code and data are publicly available to ensure reproducibility and to support the development of robust forensic detectors to keep pace with the rise of new synthetic generators.

Davide Maltoni, Matteo Ferrara

Recent large language models have demonstrated relevant capabilities in solving problems that require logical reasoning; however, the corresponding internal mechanisms remain largely unexplored. In this paper, we show that a small language model can solve a deductive reasoning task by learning the underlying rules (rather than operating as a statistical learner). A low-level explanation of its internal representations and computational circuits is then provided. Our findings reveal that induction heads play a central role in the implementation of the rule completion and rule chaining steps involved in the logical inference required by the task.

Davide Maltoni, Lorenzo Pellegrini

TPC (Three-Phase Consolidation) is here introduced as a simple but effective approach to continually learn new classes (and/or instances of known classes) while controlling forgetting of previous knowledge. Each experience (a.k.a. task) is learned in three phases characterized by different rules and learning dynamics, aimed at removing the class-bias problem (due to class unbalancing) and limiting gradient-based corrections to prevent forgetting of underrepresented classes. Several experiments on complex datasets demonstrate its accuracy and efficiency advantages over competitive existing approaches. The algorithm and all the results presented in this paper are fully reproducible thanks to its publication on the Avalanche open framework for continual learning.

Lorenzo Vorabbi, Davide Maltoni, Guido Borghi et al.

On-device learning remains a formidable challenge, especially when dealing with resource-constrained devices that have limited computational capabilities. This challenge is primarily rooted in two key issues: first, the memory available on embedded devices is typically insufficient to accommodate the memory-intensive back-propagation algorithm, which often relies on floating-point precision. Second, the development of learning algorithms on models with extreme quantization levels, such as Binary Neural Networks (BNNs), is critical due to the drastic reduction in bit representation. In this study, we propose a solution that combines recent advancements in the field of Continual Learning (CL) and Binary Neural Networks to enable on-device training while maintaining competitive performance. Specifically, our approach leverages binary latent replay (LR) activations and a novel quantization scheme that significantly reduces the number of bits required for gradient computation. The experimental validation demonstrates a significant accuracy improvement in combination with a noticeable reduction in memory requirement, confirming the suitability of our approach in expanding the practical applications of deep learning in real-world scenarios.

Lorenzo Vorabbi, Davide Maltoni, Stefano Santi

Binary Neural Networks (BNNs) use 1-bit weights and activations to efficiently execute deep convolutional neural networks on edge devices. Nevertheless, the binarization of the first layer is conventionally excluded, as it leads to a large accuracy loss. The few works addressing the first layer binarization, typically increase the number of input channels to enhance data representation; such data expansion raises the amount of operations needed and it is feasible only on systems with enough computational resources. In this work, we present a new method to binarize the first layer using directly the 8-bit representation of input data; we exploit the standard bit-planes encoding to extract features bit-wise (using depth-wise convolutions); after a re-weighting stage, features are fused again. The resulting model is fully binarized and our first layer binarization approach is model independent. The concept is evaluated on three classification datasets (CIFAR10, SVHN and CIFAR100) for different model architectures (VGG and ResNet) and, the proposed technique outperforms state of the art methods both in accuracy and BMACs reduction.

Andrea Cossu, Gabriele Graffieti, Lorenzo Pellegrini et al.

The ability of a model to learn continually can be empirically assessed in different continual learning scenarios. Each scenario defines the constraints and the opportunities of the learning environment. Here, we challenge the current trend in the continual learning literature to experiment mainly on class-incremental scenarios, where classes present in one experience are never revisited. We posit that an excessive focus on this setting may be limiting for future research on continual learning, since class-incremental scenarios artificially exacerbate catastrophic forgetting, at the expense of other important objectives like forward transfer and computational efficiency. In many real-world environments, in fact, repetition of previously encountered concepts occurs naturally and contributes to softening the disruption of previous knowledge. We advocate for a more in-depth study of alternative continual learning scenarios, in which repetition is integrated by design in the stream of incoming information. Starting from already existing proposals, we describe the advantages such class-incremental with repetition scenarios could offer for a more comprehensive assessment of continual learning models.

Lorenzo Pellegrini, Vincenzo Lomonaco, Gabriele Graffieti et al.

On-device training for personalized learning is a challenging research problem. Being able to quickly adapt deep prediction models at the edge is necessary to better suit personal user needs. However, adaptation on the edge poses some questions on both the efficiency and sustainability of the learning process and on the ability to work under shifting data distributions. Indeed, naively fine-tuning a prediction model only on the newly available data results in catastrophic forgetting, a sudden erasure of previously acquired knowledge. In this paper, we detail the implementation and deployment of a hybrid continual learning strategy (AR1*) on a native Android application for real-time on-device personalization without forgetting. Our benchmark, based on an extension of the CORe50 dataset, shows the efficiency and effectiveness of our solution.

Vincenzo Lomonaco, Lorenzo Pellegrini, Pau Rodriguez et al.

In the last few years, we have witnessed a renewed and fast-growing interest in continual learning with deep neural networks with the shared objective of making current AI systems more adaptive, efficient and autonomous. However, despite the significant and undoubted progress of the field in addressing the issue of catastrophic forgetting, benchmarking different continual learning approaches is a difficult task by itself. In fact, given the proliferation of different settings, training and evaluation protocols, metrics and nomenclature, it is often tricky to properly characterize a continual learning algorithm, relate it to other solutions and gauge its real-world applicability. The first Continual Learning in Computer Vision challenge held at CVPR in 2020 has been one of the first opportunities to evaluate different continual learning algorithms on a common hardware with a large set of shared evaluation metrics and 3 different settings based on the realistic CORe50 video benchmark. In this paper, we report the main results of the competition, which counted more than 79 teams registered, 11 finalists and 2300$ in prizes. We also summarize the winning approaches, current challenges and future research directions.

Kiran Raja, Matteo Ferrara, Annalisa Franco et al.

Morphing attacks have posed a severe threat to Face Recognition System (FRS). Despite the number of advancements reported in recent works, we note serious open issues such as independent benchmarking, generalizability challenges and considerations to age, gender, ethnicity that are inadequately addressed. Morphing Attack Detection (MAD) algorithms often are prone to generalization challenges as they are database dependent. The existing databases, mostly of semi-public nature, lack in diversity in terms of ethnicity, various morphing process and post-processing pipelines. Further, they do not reflect a realistic operational scenario for Automated Border Control (ABC) and do not provide a basis to test MAD on unseen data, in order to benchmark the robustness of algorithms. In this work, we present a new sequestered dataset for facilitating the advancements of MAD where the algorithms can be tested on unseen data in an effort to better generalize. The newly constructed dataset consists of facial images from 150 subjects from various ethnicities, age-groups and both genders. In order to challenge the existing MAD algorithms, the morphed images are with careful subject pre-selection created from the contributing images, and further post-processed to remove morphing artifacts. The images are also printed and scanned to remove all digital cues and to simulate a realistic challenge for MAD algorithms. Further, we present a new online evaluation platform to test algorithms on sequestered data. With the platform we can benchmark the morph detection performance and study the generalization ability. This work also presents a detailed analysis on various subsets of sequestered data and outlines open challenges for future directions in MAD research.

Qi She, Fan Feng, Qi Liu et al.

This report summarizes IROS 2019-Lifelong Robotic Vision Competition (Lifelong Object Recognition Challenge) with methods and results from the top $8$ finalists (out of over~$150$ teams). The competition dataset (L)ifel(O)ng (R)obotic V(IS)ion (OpenLORIS) - Object Recognition (OpenLORIS-object) is designed for driving lifelong/continual learning research and application in robotic vision domain, with everyday objects in home, office, campus, and mall scenarios. The dataset explicitly quantifies the variants of illumination, object occlusion, object size, camera-object distance/angles, and clutter information. Rules are designed to quantify the learning capability of the robotic vision system when faced with the objects appearing in the dynamic environments in the contest. Individual reports, dataset information, rules, and released source code can be found at the project homepage: "https://lifelong-robotic-vision.github.io/competition/".

Lorenzo Pellegrini, Gabriele Graffieti, Vincenzo Lomonaco et al.

Training deep neural networks at the edge on light computational devices, embedded systems and robotic platforms is nowadays very challenging. Continual learning techniques, where complex models are incrementally trained on small batches of new data, can make the learning problem tractable even for CPU-only embedded devices enabling remarkable levels of adaptiveness and autonomy. However, a number of practical problems need to be solved: catastrophic forgetting before anything else. In this paper we introduce an original technique named "Latent Replay" where, instead of storing a portion of past data in the input space, we store activations volumes at some intermediate layer. This can significantly reduce the computation and storage required by native rehearsal. To keep the representation stable and the stored activations valid we propose to slow-down learning at all the layers below the latent replay one, leaving the layers above free to learn at full pace. In our experiments we show that Latent Replay, combined with existing continual learning techniques, achieves state-of-the-art performance on complex video benchmarks such as CORe50 NICv2 (with nearly 400 small and highly non-i.i.d. batches) and OpenLORIS. Finally, we demonstrate the feasibility of nearly real-time continual learning on the edge through the deployment of the proposed technique on a smartphone device.

Vincenzo Lomonaco, Davide Maltoni, Lorenzo Pellegrini

Robotic vision is a field where continual learning can play a significant role. An embodied agent operating in a complex environment subject to frequent and unpredictable changes is required to learn and adapt continuously. In the context of object recognition, for example, a robot should be able to learn (without forgetting) objects of never before seen classes as well as improving its recognition capabilities as new instances of already known classes are discovered. Ideally, continual learning should be triggered by the availability of short videos of single objects and performed on-line on on-board hardware with fine-grained updates. In this paper, we introduce a novel continual learning protocol based on the CORe50 benchmark and propose two rehearsal-free continual learning techniques, CWR* and AR1*, that can learn effectively even in the challenging case of nearly 400 small non-i.i.d. incremental batches. In particular, our experiments show that AR1* can outperform other state-of-the-art rehearsal-free techniques by more than 15% accuracy in some cases, with a very light and constant computational and memory overhead across training batches.

Timothée Lesort, Vincenzo Lomonaco, Andrei Stoian et al.

Continual learning (CL) is a particular machine learning paradigm where the data distribution and learning objective changes through time, or where all the training data and objective criteria are never available at once. The evolution of the learning process is modeled by a sequence of learning experiences where the goal is to be able to learn new skills all along the sequence without forgetting what has been previously learned. Continual learning also aims at the same time at optimizing the memory, the computation power and the speed during the learning process. An important challenge for machine learning is not necessarily finding solutions that work in the real world but rather finding stable algorithms that can learn in real world. Hence, the ideal approach would be tackling the real world in a embodied platform: an autonomous agent. Continual learning would then be effective in an autonomous agent or robot, which would learn autonomously through time about the external world, and incrementally develop a set of complex skills and knowledge. Robotic agents have to learn to adapt and interact with their environment using a continuous stream of observations. Some recent approaches aim at tackling continual learning for robotics, but most recent papers on continual learning only experiment approaches in simulation or with static datasets. Unfortunately, the evaluation of those algorithms does not provide insights on whether their solutions may help continual learning in the context of robotics. This paper aims at reviewing the existing state of the art of continual learning, summarizing existing benchmarks and metrics, and proposing a framework for presenting and evaluating both robotics and non robotics approaches in a way that makes transfer between both fields easier.

Vincenzo Lomonaco, Karan Desai, Eugenio Culurciello et al.

High-dimensional always-changing environments constitute a hard challenge for current reinforcement learning techniques. Artificial agents, nowadays, are often trained off-line in very static and controlled conditions in simulation such that training observations can be thought as sampled i.i.d. from the entire observations space. However, in real world settings, the environment is often non-stationary and subject to unpredictable, frequent changes. In this paper we propose and openly release CRLMaze, a new benchmark for learning continually through reinforcement in a complex 3D non-stationary task based on ViZDoom and subject to several environmental changes. Then, we introduce an end-to-end model-free continual reinforcement learning strategy showing competitive results with respect to four different baselines and not requiring any access to additional supervised signals, previously encountered environmental conditions or observations.

Matteo Ferrara, Annalisa Franco, Davide Maltoni

Face morphing represents nowadays a big security threat in the context of electronic identity documents as well as an interesting challenge for researchers in the field of face recognition. Despite of the good performance obtained by state-of-the-art approaches on digital images, no satisfactory solutions have been identified so far to deal with cross-database testing and printed-scanned images (typically used in many countries for document issuing). In this work, novel approaches are proposed to train Deep Neural Networks for morphing detection: in particular generation of simulated printed-scanned images together with other data augmentation strategies and pre-training on large face recognition datasets, allowed to reach state-of-the-art accuracy on challenging datasets from heterogeneous image sources.

Natalia Díaz-Rodríguez, Vincenzo Lomonaco, David Filliat et al.

Continual learning consists of algorithms that learn from a stream of data/tasks continuously and adaptively thought time, enabling the incremental development of ever more complex knowledge and skills. The lack of consensus in evaluating continual learning algorithms and the almost exclusive focus on forgetting motivate us to propose a more comprehensive set of implementation independent metrics accounting for several factors we believe have practical implications worth considering in the deployment of real AI systems that learn continually: accuracy or performance over time, backward and forward knowledge transfer, memory overhead as well as computational efficiency. Drawing inspiration from the standard Multi-Attribute Value Theory (MAVT) we further propose to fuse these metrics into a single score for ranking purposes and we evaluate our proposal with five continual learning strategies on the iCIFAR-100 continual learning benchmark.

Davide Maltoni, Vincenzo Lomonaco

It was recently shown that architectural, regularization and rehearsal strategies can be used to train deep models sequentially on a number of disjoint tasks without forgetting previously acquired knowledge. However, these strategies are still unsatisfactory if the tasks are not disjoint but constitute a single incremental task (e.g., class-incremental learning). In this paper we point out the differences between multi-task and single-incremental-task scenarios and show that well-known approaches such as LWF, EWC and SI are not ideal for incremental task scenarios. A new approach, denoted as AR1, combining architectural and regularization strategies is then specifically proposed. AR1 overhead (in term of memory and computation) is very small thus making it suitable for online learning. When tested on CORe50 and iCIFAR-100, AR1 outperformed existing regularization strategies by a good margin.

Vincenzo Lomonaco, Davide Maltoni

Continuous/Lifelong learning of high-dimensional data streams is a challenging research problem. In fact, fully retraining models each time new data become available is infeasible, due to computational and storage issues, while naïve incremental strategies have been shown to suffer from catastrophic forgetting. In the context of real-world object recognition applications (e.g., robotic vision), where continuous learning is crucial, very few datasets and benchmarks are available to evaluate and compare emerging techniques. In this work we propose a new dataset and benchmark CORe50, specifically designed for continuous object recognition, and introduce baseline approaches for different continuous learning scenarios.

Davide Maltoni, Vincenzo Lomonaco

Recent works demonstrated the usefulness of temporal coherence to regularize supervised training or to learn invariant features with deep architectures. In particular, enforcing smooth output changes while presenting temporally-closed frames from video sequences, proved to be an effective strategy. In this paper we prove the efficacy of temporal coherence for semi-supervised incremental tuning. We show that a deep architecture, just mildly trained in a supervised manner, can progressively improve its classification accuracy, if exposed to video sequences of unlabeled data. The extent to which, in some cases, a semi-supervised tuning allows to improve classification accuracy (approaching the supervised one) is somewhat surprising. A number of control experiments pointed out the fundamental role of temporal coherence.