Effectively learning from sequential data is a longstanding goal of Artificial Intelligence, especially in the case of long sequences. From the dawn of Machine Learning, several researchers have pursued algorithms and architectures capable of processing sequences of patterns, retaining information about past inputs while still leveraging future data, without losing precious long-term dependencies and correlations. While such an ultimate goal is inspired by the human hallmark of continuous real-time processing of sensory information, several solutions have simplified the learning paradigm by artificially limiting the processed context or dealing with sequences of limited length, given in advance. These solutions were further emphasized by the ubiquity of Transformers, which initially overshadowed the role of Recurrent Neural Nets. However, recurrent networks are currently experiencing a strong recent revival due to the growing popularity of (deep) State-Space models and novel instances of large-context Transformers, which are both based on recurrent computations that aim to go beyond several limits of currently ubiquitous technologies. The fast development of Large Language Models has renewed the interest in efficient solutions to process data over time. This survey provides an in-depth summary of the latest approaches that are based on recurrent models for sequential data processing. A complete taxonomy of recent trends in architectural and algorithmic solutions is reported and discussed, guiding researchers in this appealing research field. The emerging picture suggests that there is room for exploring novel routes, constituted by learning algorithms that depart from the standard Backpropagation Through Time, towards a more realistic scenario where patterns are effectively processed online, leveraging local-forward computations, and opening new directions for research on this topic.

The growing ubiquity of Retrieval-Augmented Generation (RAG) systems in several real-world services triggers severe concerns about their security. A RAG system improves the generative capabilities of a Large Language Model (LLM) by a retrieval mechanism that operates on a private knowledge base, whose unintended exposure could lead to severe consequences, including breaches of private and sensitive information. This paper presents a black-box attack to force a RAG system to leak its private knowledge base which, unlike existing approaches, is both adaptive and automatic. A relevance-based mechanism and an attacker-side open-source LLM favor the generation of effective queries to leak most of the (hidden) knowledge base. Extensive experimentation proves the quality of the proposed algorithm in different RAG pipelines and domains, compared to very recent related approaches, which turn out to be either not fully black-box, not adaptive, or not based on open-source models. The findings from our study highlight the urgent need for more robust privacy safeguards in the design and deployment of RAG systems. We have made the open-source code for our experimental procedure available for public use [12].

The notion of stability plays a crucial role in ensuring the safe development of a model in a lifelong learning context.This paper investigates the fundamental aspects of stability in a class of continuous-time recurrent neural networks which include both state and costate variables.The latter are directly inherited from optimal control theory, and they act as adjoint variables closely related to gradient terms.Stability is investigated both in terms of state and of costate dynamics, showing the key conditions that must be satisfied to produce bounded dynamics in the forward and learning stages.* This work was

One of the goals of neuro-symbolic artificial intelligence is to exploit background knowledge to improve the performance of learning tasks. However, most of the existing frameworks focus on the simplified scenario where knowledge does not change over time and does not cover the temporal dimension. In this work we consider the much more challenging problem of knowledge-driven sequence classification where different portions of knowledge must be employed at different timesteps, and temporal relations are available. Our extensive experimental evaluation compares multi-stage neuro-symbolic and neural-only architectures, and it is conducted on a newly-introduced benchmarking framework. Results not only demonstrate the challenging nature of this novel setting, but also highlight under-explored shortcomings of neuro-symbolic methods, representing a precious reference for future research.

In this study, we investigate the continuous time dynamics of Recurrent Neural Networks (RNNs), focusing on systems with nonlinear activation functions. The objective of this work is to identify conditions under which RNNs exhibit perpetual oscillatory behavior, without converging to static fixed points. We establish that skew-symmetric weight matrices are fundamental to enable stable limit cycles in both linear and nonlinear configurations. We further demonstrate that hyperbolic tangent-like activation functions (odd, bounded, and continuous) preserve these oscillatory dynamics by ensuring motion invariants in state space. Numerical simulations showcase how nonlinear activation functions not only maintain limit cycles, but also enhance the numerical stability of the system integration process, mitigating those instabilities that are commonly associated with the forward Euler method. The experimental results of this analysis highlight practical considerations for designing neural architectures capable of capturing complex temporal dependencies, i.e., strategies for enhancing memorization skills in recurrent models.

Storing and handling huge data collections has become the fundamental player in the progress of Machine Learning and of its spectacular results. However, learning from such data collections introduces risks related to data centralization, privacy, energy efficiency, limited customizability, and control. This paper sustains the position that the time has come for thinking of new learning protocols where machines conquer cognitive skills by online learning from potentially lifelong streams of sensory data, without the privilege of recording the temporal stream. The perspective of what we refer to as "Collectionless AI" pushes towards interactions with the environment, including humans and other artificial agents, to favor dynamic adaptation, customizability, control. At each time instant, data acquired from the environment is only processed with the purpose of contributing to update the current agent-internal representation of the environment, promoting the development of self-organized memorization skills. The goal of this paper is not to introduce new algorithms, but to present an extreme perspective which recovers largely known notions out of the current mainstream, with the goal of stimulating the development of new foundations on computational processes of learning and reasoning. This might open the doors to a truly orthogonal competitive track on AI technologies that avoid data accumulation by design, thus offering a framework which is better suited concerning privacy issues, control and customizability. Pushing towards massively distributed computation, the collectionless approach to AI might reduce the concentration of power in companies and governments, better facing geopolitical issues.

Nowadays, services exploiting Large Language Models (LLMs) frequently customize the model responses in function of their specific requirements, thanks to Retrieval-Augmented Generation (RAG). While this is a simple and effective solution, it has been recently shown to suffer from serious LLM-related security issues that, in some circumstances, can be exploited to access the RAG internal knowledge base, possibly containing private/sensitive data. This paper (i) compares different existing recent attacks to RAG systems in three real-world scenarios, with the goal of in-depth evaluating them. Moreover, novel attacks are proposed, in order to study distinct open directions in this field of study: (ii) the first one is based on free-online-API variants of recent black-box attacks, which can be run on a domestic machine, and a fully black-box reformulation of a gray-box method, in both cases simulating more realistic conditions built with open-source tools; (iii) the second one is aimed at building automatic procedures to get the most out of the knowledge-base of a RAG system, introducing a strategy that allows a recent memory-based approach to become automatic. A large experimental comparison is included that, to our best knowledge, is the most extended one in the current scientific literature. Results highlight the sensitivity of the attack procedure to the basic tools over which they are built, but also shows that open-source solutions can be easily exploited to setup attacks. Moreover, the actually feasibility of designing an automatic procedure is proved. In both the cases, this paper raises an important warning on the need of taking specific precautions to setup robust RAG systems.

Learning with neural networks from a continuous stream of visual information presents several challenges due to the non-i.i.d. nature of the data. However, it also offers novel opportunities to develop representations that are consistent with the information flow. In this paper we investigate the case of unsupervised continual learning of pixel-wise features subject to multiple motion-induced constraints, therefore named motion-conjugated feature representations. Differently from existing approaches, motion is not a given signal (either ground-truth or estimated by external modules), but is the outcome of a progressive and autonomous learning process, occurring at various levels of the feature hierarchy. Multiple motion flows are estimated with neural networks and characterized by different levels of abstractions, spanning from traditional optical flow to other latent signals originating from higher-level features, hence called higher-order motions. Continuously learning to develop consistent multi-order flows and representations is prone to trivial solutions, which we counteract by introducing a self-supervised contrastive loss, spatially-aware and based on flow-induced similarity. We assess our model on photorealistic synthetic streams and real-world videos, comparing to pre-trained state-of-the art feature extractors (also based on Transformers) and to recent unsupervised learning models, significantly outperforming these alternatives.

Optimal control deals with optimization problems in which variables steer a dynamical system, and its outcome contributes to the objective function. Two classical approaches to solving these problems are Dynamic Programming and the Pontryagin Maximum Principle. In both approaches, Hamiltonian equations offer an interpretation of optimality through auxiliary variables known as costates. However, Hamiltonian equations are rarely used due to their reliance on forward-backward algorithms across the entire temporal domain. This paper introduces a novel neural-based approach to optimal control. Neural networks are employed not only for implementing state dynamics but also for estimating costate variables. The parameters of the latter network are determined at each time step using a newly introduced local policy referred to as the time-reversed generalized Riccati equation. This policy is inspired by a result discussed in the Linear Quadratic (LQ) problem, which we conjecture stabilizes state dynamics. We support this conjecture by discussing experimental results from a range of optimal control case studies.

Efficiently learning unsupervised pixel-wise visual representations is crucial for training agents that can perceive their environment without relying on heavy human supervision or abundant annotated data. Motivated by recent work that promotes motion as a key source of information in representation learning, we propose a novel instance of contrastive criterions over time and space. In our architecture, pixel-wise motion field and representations are extracted by neural models, trained from scratch in an integrated fashion. Learning proceeds online over time, exploiting also a momentum-based moving average to update the feature extractor, without replaying any large buffers of past data. Experiments on real-world videos and on a recently introduced benchmark, with photorealistic streams generated from a 3D environment, confirm that the proposed model can learn to estimate motion and jointly develop representations. Our model nicely encodes the variable appearance of the visual information in space and time, significantly overcoming a recent approach and it also compares favourably with convolutional and Transformer-based networks, offline-pre-trained on large collections of supervised and unsupervised images.

The intrinsic difficulty in adapting deep learning models to non-stationary environments limits the applicability of neural networks to real-world tasks. This issue is critical in practical supervised learning settings, such as the ones in which a pre-trained model computes projections toward a latent space where different task predictors are sequentially learned over time. As a matter of fact, incrementally fine-tuning the whole model to better adapt to new tasks usually results in catastrophic forgetting, with decreasing performance over the past experiences and losing valuable knowledge from the pretraining stage. In this paper, we propose a novel strategy to make the fine-tuning procedure more effective, by avoiding to update the pre-trained part of the network and learning not only the usual classification head, but also a set of newly-introduced learnable parameters that are responsible for transforming the input data. This process allows the network to effectively leverage the pre-training knowledge and find a good trade-off between plasticity and stability with modest computational efforts, thus especially suitable for on-the-edge settings. Our experiments on four image classification problems in a continual learning setting confirm the quality of the proposed approach when compared to several fine-tuning procedures and to popular continual learning methods.

Fast reactions to changes in the surrounding visual environment require efficient attention mechanisms to reallocate computational resources to most relevant locations in the visual field. While current computational models keep improving their predictive ability thanks to the increasing availability of data, they still struggle approximating the effectiveness and efficiency exhibited by foveated animals. In this paper, we present a biologically-plausible computational model of focus of attention that exhibits spatiotemporal locality and that is very well-suited for parallel and distributed implementations. Attention emerges as a wave propagation process originated by visual stimuli corresponding to details and motion information. The resulting field obeys the principle of "inhibition of return" so as not to get stuck in potential holes. An accurate experimentation of the model shows that it achieves top level performance in scanpath prediction tasks. This can easily be understood at the light of a theoretical result that we establish in the paper, where we prove that as the velocity of wave propagation goes to infinity, the proposed model reduces to recently proposed state of the art gravitational models of focus of attention.

Explainable artificial intelligence has rapidly emerged since lawmakers have started requiring interpretable models for safety-critical domains. Concept-based neural networks have arisen as explainable-by-design methods as they leverage human-understandable symbols (i.e. concepts) to predict class memberships. However, most of these approaches focus on the identification of the most relevant concepts but do not provide concise, formal explanations of how such concepts are leveraged by the classifier to make predictions. In this paper, we propose a novel end-to-end differentiable approach enabling the extraction of logic explanations from neural networks using the formalism of First-Order Logic. The method relies on an entropy-based criterion which automatically identifies the most relevant concepts. We consider four different case studies to demonstrate that: (i) this entropy-based criterion enables the distillation of concise logic explanations in safety-critical domains from clinical data to computer vision; (ii) the proposed approach outperforms state-of-the-art white-box models in terms of classification accuracy.

In this paper, we present PARTIME, a software library written in Python and based on PyTorch, designed specifically to speed up neural networks whenever data is continuously streamed over time, for both learning and inference. Existing libraries are designed to exploit data-level parallelism, assuming that samples are batched, a condition that is not naturally met in applications that are based on streamed data. Differently, PARTIME starts processing each data sample at the time in which it becomes available from the stream. PARTIME wraps the code that implements a feed-forward multi-layer network and it distributes the layer-wise processing among multiple devices, such as Graphics Processing Units (GPUs). Thanks to its pipeline-based computational scheme, PARTIME allows the devices to perform computations in parallel. At inference time this results in scaling capabilities that are theoretically linear with respect to the number of devices. During the learning stage, PARTIME can leverage the non-i.i.d. nature of the streamed data with samples that are smoothly evolving over time for efficient gradient computations. Experiments are performed in order to empirically compare PARTIME with classic non-parallel neural computations in online learning, distributing operations on up to 8 NVIDIA GPUs, showing significant speedups that are almost linear in the number of devices, mitigating the impact of the data transfer overhead.

Devising intelligent agents able to live in an environment and learn by observing the surroundings is a longstanding goal of Artificial Intelligence. From a bare Machine Learning perspective, challenges arise when the agent is prevented from leveraging large fully-annotated dataset, but rather the interactions with supervisory signals are sparsely distributed over space and time. This paper proposes a novel neural-network-based approach to progressively and autonomously develop pixel-wise representations in a video stream. The proposed method is based on a human-like attention mechanism that allows the agent to learn by observing what is moving in the attended locations. Spatio-temporal stochastic coherence along the attention trajectory, paired with a contrastive term, leads to an unsupervised learning criterion that naturally copes with the considered setting. Differently from most existing works, the learned representations are used in open-set class-incremental classification of each frame pixel, relying on few supervisions. Our experiments leverage 3D virtual environments and they show that the proposed agents can learn to distinguish objects just by observing the video stream. Inheriting features from state-of-the art models is not as powerful as one might expect.

Deploying AI-powered systems requires trustworthy models supporting effective human interactions, going beyond raw prediction accuracy. Concept bottleneck models promote trustworthiness by conditioning classification tasks on an intermediate level of human-like concepts. This enables human interventions which can correct mispredicted concepts to improve the model's performance. However, existing concept bottleneck models are unable to find optimal compromises between high task accuracy, robust concept-based explanations, and effective interventions on concepts---particularly in real-world conditions where complete and accurate concept supervisions are scarce. To address this, we propose Concept Embedding Models, a novel family of concept bottleneck models which goes beyond the current accuracy-vs-interpretability trade-off by learning interpretable high-dimensional concept representations. Our experiments demonstrate that Concept Embedding Models (1) attain better or competitive task accuracy w.r.t. standard neural models without concepts, (2) provide concept representations capturing meaningful semantics including and beyond their ground truth labels, (3) support test-time concept interventions whose effect in test accuracy surpasses that in standard concept bottleneck models, and (4) scale to real-world conditions where complete concept supervisions are scarce.

Adversarial attacks on machine learning-based classifiers, along with defense mechanisms, have been widely studied in the context of single-label classification problems. In this paper, we shift the attention to multi-label classification, where the availability of domain knowledge on the relationships among the considered classes may offer a natural way to spot incoherent predictions, i.e., predictions associated to adversarial examples lying outside of the training data distribution. We explore this intuition in a framework in which first-order logic knowledge is converted into constraints and injected into a semi-supervised learning problem. Within this setting, the constrained classifier learns to fulfill the domain knowledge over the marginal distribution, and can naturally reject samples with incoherent predictions. Even though our method does not exploit any knowledge of attacks during training, our experimental analysis surprisingly unveils that domain-knowledge constraints can help detect adversarial examples effectively, especially if such constraints are not known to the attacker. We show how to implement an adaptive attack exploiting knowledge of the constraints and, in a specifically-designed setting, we provide experimental comparisons with popular state-of-the-art attacks. We believe that our approach may provide a significant step towards designing more robust multi-label classifiers.

The popularity of deep learning techniques renewed the interest in neural architectures able to process complex structures that can be represented using graphs, inspired by Graph Neural Networks (GNNs). We focus our attention on the originally proposed GNN model of Scarselli et al. 2009, which encodes the state of the nodes of the graph by means of an iterative diffusion procedure that, during the learning stage, must be computed at every epoch, until the fixed point of a learnable state transition function is reached, propagating the information among the neighbouring nodes. We propose a novel approach to learning in GNNs, based on constrained optimization in the Lagrangian framework. Learning both the transition function and the node states is the outcome of a joint process, in which the state convergence procedure is implicitly expressed by a constraint satisfaction mechanism, avoiding iterative epoch-wise procedures and the network unfolding. Our computational structure searches for saddle points of the Lagrangian in the adjoint space composed of weights, nodes state variables and Lagrange multipliers. This process is further enhanced by multiple layers of constraints that accelerate the diffusion process. An experimental analysis shows that the proposed approach compares favourably with popular models on several benchmarks.

In the last few years, the scientific community showed a remarkable and increasing interest towards 3D Virtual Environments, training and testing Machine Learning-based models in realistic virtual worlds. On one hand, these environments could also become a mean to study the weaknesses of Machine Learning algorithms, or to simulate training settings that allow Machine Learning models to gain robustness to 3D adversarial attacks. On the other hand, their growing popularity might also attract those that aim at creating adversarial conditions to invalidate the benchmarking process, especially in the case of public environments that allow the contribution from a large community of people. Most of the existing Adversarial Machine Learning approaches are focused on static images, and little work has been done in studying how to deal with 3D environments and how a 3D object should be altered to fool a classifier that observes it. In this paper, we study how to craft adversarial 3D objects by altering their textures, using a tool chain composed of easily accessible elements. We show that it is possible, and indeed simple, to create adversarial objects using off-the-shelf limited surrogate renderers that can compute gradients with respect to the parameters of the rendering process, and, to a certain extent, to transfer the attacks to more advanced 3D engines. We propose a saliency-based attack that intersects the two classes of renderers in order to focus the alteration to those texture elements that are estimated to be effective in the target engine, evaluating its impact in popular neural classifiers.

In the last decade, motivated by the success of Deep Learning, the scientific community proposed several approaches to make the learning procedure of Neural Networks more effective. When focussing on the way in which the training data are provided to the learning machine, we can distinguish between the classic random selection of stochastic gradient-based optimization and more involved techniques that devise curricula to organize data, and progressively increase the complexity of the training set. In this paper, we propose a novel training procedure named Friendly Training that, differently from the aforementioned approaches, involves altering the training examples in order to help the model to better fulfil its learning criterion. The model is allowed to “simplify” those examples that are too hard to be classified at a certain stage of the training procedure. The data transformation is controlled by a developmental plan that progressively reduces its impact during training, until it completely vanishes. In a sense, this is the opposite of what is commonly done in order to increase robustness against adversarial examples, i.e., Adversarial Training. Experiments on multiple datasets are provided, showing that Friendly Training yields improvements with respect to informed data sub-selection routines and random selection, especially in deep convolutional architectures. Results suggest that adapting the input data is a feasible way to stabilize learning and improve the generalization skills of the network.

Writers, poets, singers usually do not create their compositions in just one breath. Text is revisited, adjusted, modified, rephrased, even multiple times, in order to better convey meanings, emotions and feelings that the author wants to express. Amongst the noble written arts, Poetry is probably the one that needs to be elaborated the most, since the composition has to formally respect predefined meter and rhyming schemes. In this paper, we propose a framework to generate poems that are repeatedly revisited and corrected, as humans do, in order to improve their overall quality. We frame the problem of revising poems in the context of Reinforcement Learning and, in particular, using Proximal Policy Optimization. Our model generates poems from scratch and it learns to progressively adjust the generated text in order to match a target criterion. We evaluate this approach in the case of matching a rhyming scheme, without having any information on which words are responsible of creating rhymes and on how to coherently alter the poem words. The proposed framework is general and, with an appropriate reward shaping, it can be applied to other text generation problems.

In the last few years we have seen a remarkable progress from the cultivation of the idea of expressing domain knowledge by the mathematical notion of constraint. However, the progress has mostly involved the process of providing consistent solutions with a given set of constraints, whereas learning “new” constraints, that express new knowledge, is still an open challenge. In this paper we propose a novel approach to learning of constraints which is based on information theoretic principles. The basic idea consists in maximizing the transfer of information between task functions and a set of learnable constraints, implemented using neural networks subject to L1 regularization. This process leads to the unsupervised development of new constraints that are fulfilled in different sub-portions of the input domain. In addition, we define a simple procedure that can explain the behaviour of the newly devised constraints in terms of First-Order Logic formulas, thus extracting novel knowledge on the relationships between the original tasks. An experimental evaluation is provided to support the proposed approach, in which we also explore the regularization effects introduced by the proposed Information-Based Learning of Constraint (IBLC) algorithm.

In many real world applications, data are characterized by a complex structure, that can be naturally encoded as a graph.In the last years, the popularity of deep learning techniques has renewed the interest in neural models able to process complex patterns.In particular, inspired by the Graph Neural Network (GNN) model, different architectures have been proposed to extend the original GNN scheme.GNNs exploit a set of state variables, each assigned to a graph node, and a diffusion mechanism of the states among neighbor nodes, to implement an iterative procedure to compute the fixed point of the (learnable) state transition function.In this paper, we propose a novel approach to the state computation and the learning algorithm for GNNs, based on a constraint optimisation task solved in the Lagrangian framework.The state convergence procedure is implicitly expressed by the constraint satisfaction mechanism and does not require a separate iterative phase for each epoch of the learning procedure.In fact, the computational structure is based on the search for saddle points of the Lagrangian in the adjoint space composed of weights, neural outputs (node states), and Lagrange multipliers.The proposed approach is compared experimentally with other popular models for processing graphs.

Recently, researchers in Machine Learning algorithms, Computer Vision scientists, engineers and others, showed a growing interest in 3D simulators as a mean to artificially create experimental settings that are very close to those in the real world. However, most of the existing platforms to interface algorithms with 3D environments are often designed to setup navigation-related experiments, to study physical interactions, or to handle ad-hoc cases that are not thought to be customized, sometimes lacking a strong photorealistic appearance and an easy-to-use software interface. In this paper, we present a novel platform, SAILenv, that is specifically designed to be simple and customizable, and that allows researchers to experiment visual recognition in virtual 3D scenes. A few lines of code are needed to interface every algorithm with the virtual world, and non-3D-graphics experts can easily customize the 3D environment itself, exploiting a collection of photorealistic objects. Our framework yields pixel-level semantic and instance labeling, depth, and, to the best of our knowledge, it is the only one that provides motion-related information directly inherited from the 3D engine. The client-server communication operates at a low level, avoiding the overhead of HTTP-based data exchanges. We perform experiments using a state-of-the-art object detector trained on real-world images, showing that it is able to recognize the photorealistic 3D objects of our environment. The computational burden of the optical flow compares favourably with the estimation performed using modern GPU-based convolutional networks or more classic implementations. We believe that the scientific community will benefit from the easiness and high-quality of our framework to evaluate newly proposed algorithms in their own customized realistic conditions.

Deep neural networks are usually considered black-boxes due to their complex internal architecture, that cannot straightforwardly provide human-understandable explanations on how they behave. Indeed, Deep Learning is still viewed with skepticism in those real-world domains in which incorrect predictions may produce critical effects. This is one of the reasons why in the last few years Explainable Artificial Intelligence (XAI) techniques have gained a lot of attention in the scientific community. In this paper, we focus on the case of multi-label classification, proposing a neural network that learns the relationships among the predictors associated to each class, yielding First-Order Logic (FOL)-based descriptions. Both the explanation-related network and the classification-related network are jointly learned, thus implicitly introducing a latent dependency between the development of the explanation mechanism and the development of the classifiers. Our model can integrate human-driven preferences that guide the learning-to-explain process, and it is presented in a unified framework. Different typologies of explanations are evaluated in distinct experiments, showing that the proposed approach discovers new knowledge and can improve the classifier performance.

In this paper we present a foundational study on a constrained method that defines learning problems with Neural Networks in the context of the principle of least cognitive action, which very much resembles the principle of least action in mechanics. Starting from a general approach to enforce constraints into the dynamical laws of learning, this work focuses on an alternative way of defining Neural Networks, that is different from the majority of existing approaches. In particular, the structure of the neural architecture is defined by means of a special class of constraints that are extended also to the interaction with data, leading to "architectural" and "input-related" constraints, respectively. The proposed theory is cast into the time domain, in which data are presented to the network in an ordered manner, that makes this study an important step toward alternative ways of processing continuous streams of data with Neural Networks. The connection with the classic Backpropagation-based update rule of the weights of networks is discussed, showing that there are conditions under which our approach degenerates to Backpropagation. Moreover, the theory is experimentally evaluated on a simple problem that allows us to deeply study several aspects of the theory itself and to show the soundness of the model.

In this paper we study a constraint-based representation of neural network architectures. We cast the learning problem in the Lagrangian framework and we investigate a simple optimization procedure that is well suited to fulfil the so-called architectural constraints, learning from the available supervisions. The computational structure of the proposed Local Propagation (LP) algorithm is based on the search for saddle points in the adjoint space composed of weights, neural outputs, and Lagrange multipliers. All the updates of the model variables are locally performed, so that LP is fully parallelizable over the neural units, circumventing the classic problem of gradient vanishing in deep networks. The implementation of popular neural models is described in the context of LP, together with those conditions that trace a natural connection with Backpropagation. We also investigate the setting in which we tolerate bounded violations of the architectural constraints, and we provide experimental evidence that LP is a feasible approach to train shallow and deep networks, opening the road to further investigations on more complex architectures, easily describable by constraints.

Human visual attention is a complex phenomenon. A computational modeling of this phenomenon must take into account where people look in order to evaluate which are the salient locations (spatial distribution of the fixations), when they look in those locations to understand the temporal development of the exploration (temporal order of the fixations), and how they move from one location to another with respect to the dynamics of the scene and the mechanics of the eyes (dynamics). State-of-the-art models focus on learning saliency maps from human data, a process that only takes into account the spatial component of the phenomenon and ignore its temporal and dynamical counterparts. In this work we focus on the evaluation methodology of models of human visual attention. We underline the limits of the current metrics for saliency prediction and scanpath similarity, and we introduce a statistical measure for the evaluation of the dynamics of the simulated eye movements. While deep learning models achieve astonishing performance in saliency prediction, our analysis shows their limitations in capturing the dynamics of the process. We find that unsupervised gravitational models, despite of their simplicity, outperform all competitors. Finally, exploiting a crowd-sourcing platform, we present a study aimed at evaluating how strongly the scanpaths generated with the unsupervised gravitational models appear plausible to naive and expert human observers.

Unsupervised learning from continuous visual streams is a challenging problem that cannot be naturally and efficiently managed in the classic batch-mode setting of computation. The information stream must be carefully processed accordingly to an appropriate spatio-temporal distribution of the visual data, while most approaches of learning commonly assume uniform probability density. In this paper we focus on unsupervised learning for transferring visual information in a truly online setting by using a computational model that is inspired to the principle of least action in physics. The maximization of the mutual information is carried out by a temporal process which yields online estimation of the entropy terms. The model, which is based on second-order differential equations, maximizes the information transfer from the input to a discrete space of symbols related to the visual features of the input, whose computation is supported by hidden neurons. In order to better structure the input probability distribution, we use a human-like focus of attention model that, coherently with the information maximization model, is also based on second-order differential equations. We provide experimental results to support the theory by showing that the spatio-temporal filtering induced by the focus of attention allows the system to globally transfer more information from the input stream over the focused areas and, in some contexts, over the whole frames with respect to the unfiltered case that yields uniform probability distributions.

Humans are continuously exposed to a stream of visual data with a natural temporal structure. However, most successful computer vision algorithms work at image level, completely discarding the precious information carried by motion. In this paper, we claim that processing visual streams naturally leads to formulate the motion invariance principle, which enables the construction of a new theory of learning that originates from variational principles, just like in physics. Such principled approach is well suited for a discussion on a number of interesting questions that arise in vision, and it offers a well-posed computational scheme for the discovery of convolutional filters over the retina. Differently from traditional convolutional networks, which need massive supervision, the proposed theory offers a truly new scenario for the unsupervised processing of video signals, where features are extracted in a multi-layer architecture with motion invariance. While the theory enables the implementation of novel computer vision systems, it also sheds light on the role of information-based principles to drive possible biological solutions.

The understanding of the mechanisms behind focus of attention in a visual scene is a problem of great interest in visual perception and computer vision. In this paper, we describe a model of scanpath as a dynamic process which can be interpreted as a variational law somehow related to mechanics, where the focus of attention is subject to a gravitational field. The distributed virtual mass that drives eye movements is associated with the presence of details and motion in the video. Unlike most current models, the proposed approach does not estimate directly the saliency map, but the prediction of eye movements allows us to integrate over time the positions of interest. The process of inhibition-of-return is also supported in the same dynamic model with the purpose of simulating fixations and saccades. The differential equations of motion of the proposed model are numerically integrated to simulate scanpaths on both images and videos. Experimental results for the tasks of saliency and scanpath prediction on a wide collection of datasets are presented to support the theory. Top level performances are achieved especially in the prediction of scanpaths, which is the primary purpose of the proposed model.

In this brief, the extension of the framework of Learning from Constraints (LfC) to a distributed setting where multiple parties, connected over the network, contribute to the learning process is studied. LfC relies on the generic notion of "constraint" to inject knowledge into the learning problem, and, due to its generality, it deals with possibly nonconvex constraints, enforced either in a hard or soft way. Motivated by recent progresses in the field of distributed and constrained nonconvex optimization, we apply the (distributed) asynchronous method of multipliers (ASYMM) to LfC. The study shows that such a method allows us to support scenarios where selected constraints (i.e., knowledge), data, and outcomes of the learning process can be locally stored in each computational node without being shared with the rest of the network, opening the road to further investigations into privacy-preserving LfC. Constraints act as a bridge between what is shared over the net and what is private to each node, and no central authority is required. We demonstrate the applicability of these ideas in two distributed real-world settings in the context of digit recognition and document classification.

We consider a scenario where an artificial agent is reading a stream of text composed of a set of narrations, and it is informed about the identity of some of the individuals that are mentioned in the text portion that is currently being read. The agent is expected to learn to follow the narrations, thus disambiguating mentions and discovering new individuals. We focus on the case in which individuals are entities and relations and propose an end-to-end trainable memory network that learns to discover and disambiguate them in an online manner, performing one-shot learning and dealing with a small number of sparse supervisions. Our system builds a not-given-in-advance knowledge base, and it improves its skills while reading the unsupervised text. The model deals with abrupt changes in the narration, considering their effects when resolving coreferences. We showcase the strong disambiguation and discovery skills of our model on a corpus of Wikipedia documents and on a newly introduced data set that we make publicly available.

The puzzle of computer vision might find new challenging solutions when we realize that most successful methods are working at image level, which is remarkably more difficult than processing directly visual streams, just as it happens in nature. In this paper, we claim that the processing of a stream of frames naturally leads to formulate the motion invariance principle, which enables the construction of a new theory of visual learning based on convolutional features. The theory addresses a number of intriguing questions that arise in natural vision, and offers a well-posed computational scheme for the discovery of convolutional filters over the retina. They are driven by the Euler- Lagrange differential equations derived from the principle of least cognitive action, that parallels the laws of mechanics. Unlike traditional convolutional networks, which need massive supervision, the proposed theory offers a truly new scenario in which feature learning takes place by unsupervised processing of video signals. An experimental report of the theory is presented where we show that features extracted under motion invariance yield an improvement that can be assessed by measuring information-based indexes.

This issue marks the first anniversary issue since I was honored to serve as the Editor-in-Chief (EiC) of the IEEE Transactions on Neural Networks and Learning Systems (TNNLS). I am happy to report that we had a very successful year and here are a few highlights that I would like to share with the community.•The latest impact factor of TNNLS is 4.854 according to the Journal Citation Reports. This marks a record high impact factor for our journal and places TNNLS as the number one scholarly publication in Computer Science (Hardware & Architecture), number three in Computer Science (Theory & Methods), and number ten in Electrical and Electronic Engineering journals.

The mathematical foundations of a new theory for the design of intelligent agents are presented. The proposed learning paradigm is centered around the concept of constraint, representing the interactions with the environment, and the parsimony principle. The classical regularization framework of kernel machines is naturally extended to the case in which the agents interact with a richer environment, where abstract granules of knowledge, compactly described by different linguistic formalisms, can be translated into the unified notion of constraint for defining the hypothesis set. Constrained variational calculus is exploited to derive general representation theorems that provide a description of the optimal body of the agent (i.e., the functional structure of the optimal solution to the learning problem), which is the basis for devising new learning algorithms. We show that regardless of the kind of constraints, the optimal body of the agent is a support constraint machine (SCM) based on representer theorems that extend classical results for kernel machines and provide new representations. In a sense, the expressiveness of constraints yields a semantic-based regularization theory, which strongly restricts the hypothesis set of classical regularization. Some guidelines to unify continuous and discrete computational mechanisms are given so as to accommodate in the same framework various kinds of stimuli, for example, supervised examples and logic predicates. The proposed view of learning from constraints incorporates classical learning from examples and extends naturally to the case in which the examples are subsets of the input space, which is related to learning propositional logic clauses.

A learning paradigm is proposed and investigated, in which the classical framework of learning from examples is enhanced by the introduction of hard pointwise constraints, i.e., constraints imposed on a finite set of examples that cannot be violated. Such constraints arise, e.g., when requiring coherent decisions of classifiers acting on different views of the same pattern. The classical examples of supervised learning, which can be violated at the cost of some penalization (quantified by the choice of a suitable loss function) play the role of soft pointwise constraints. Constrained variational calculus is exploited to derive a representer theorem that provides a description of the functional structure of the optimal solution to the proposed learning paradigm. It is shown that such an optimal solution can be represented in terms of a set of support constraints, which generalize the concept of support vectors and open the doors to a novel learning paradigm, called support constraint machines. The general theory is applied to derive the representation of the optimal solution to the problem of learning from hard linear pointwise constraints combined with soft pointwise constraints induced by supervised examples. In some cases, closed-form optimal solutions are obtained.

Supervised examples and prior knowledge on regions of the input space have been profitably integrated in kernel machines to improve the performance of classifiers in different real-world contexts. The proposed solutions, which rely on the unified supervision of points and sets, have been mostly based on specific optimization schemes in which, as usual, the kernel function operates on points only. In this paper, arguments from variational calculus are used to support the choice of a special class of kernels, referred to as box kernels, which emerges directly from the choice of the kernel function associated with a regularization operator. It is proven that there is no need to search for kernels to incorporate the structure deriving from the supervision of regions of the input space, because the optimal kernel arises as a consequence of the chosen regularization operator. Although most of the given results hold for sets, we focus attention on boxes, whose labeling is associated with their propositional description. Based on different assumptions, some representer theorems are given that dictate the structure of the solution in terms of box kernel expansion. Successful results are given for problems of medical diagnosis, image, and text categorization.

Based on a recently proposed framework of learning from constraints using kernel-based representations, in this brief, we naturally extend its application to the case of inferences on new constraints. We give examples for polynomials and first-order logic by showing how new constraints can be checked on the basis of given premises and data samples. Interestingly, this gives rise to a perceptual logic scheme in which the inference mechanisms do not rely only on formal schemes, but also on the data probability distribution. It is claimed that when using a properly relaxed computational checking approach, the complementary role of data samples makes it possible to break the complexity barriers of related formal checking mechanisms.

Following basic principles of information-theoretic learning, in this paper, we propose a novel approach to data clustering, referred to as minimal entropy encoding (MEE), which is based on a set of functions (features) projecting each input onto a minimum entropy configuration (code). Inspired by traditional parsimony principles, we seek solutions in reproducing kernel Hilbert spaces and then we prove that the encoding functions are expressed in terms of kernel expansion. In order to avoid trivial solutions, the developed features must be as different as possible by means of a soft constraint on the empirical estimation of the entropy associated with the encoding functions. This leads to an unconstrained optimization problem that can be efficiently solved by conjugate gradient. We also investigate an optimization strategy based on concave-convex algorithms. The relationships with maximum margin clustering are studied, showing that MEE overcomes some of its critical issues, such as the lack of a multiclass extension and the need to face problems with a large number of constraints. A massive evaluation on several benchmarks of the proposed approach shows improvements over state-of-the-art techniques, both in terms of accuracy and computational complexity.

Similarity neural networks (SNNs) are a novel neural network model designed to learn similarity measures for pairs of patterns, exploiting binary supervision. SNNs guarantee to compute non negative and symmetric measures, and show good generalization capabilities even if a small set of supervised pairs is used for training. The application of the new model to K-Means like semi-supervised clustering is investigated, introducing a technique that allows the algorithm to compute cluster centroids by means of Backpropagation on the input layer of the SNN, biased by a regularization function. The experiments carried out on some datasets from the UCI repository show that SNN based clustering almost always outperforms other methods proposed in the literature.

This paper presents an improved active shape model algorithm, that exploits auto associative neural networks (AANNs) to estimate the local feature models. The proposed technique aims at solving face feature localization tasks, nevertheless it can be used also in the more general case of object detection. Three main contributions are presented. The first one consists in the estimation of elliptic search areas by means of the training data. The second one is the use of AANNs as local feature detectors, since this network model is particularly suited to solve classification tasks with unbalanced classes. Finally, an optimized technique to set up the learning environment, needed to train the AANNs, is described. The performances of the proposed algorithm compare favorably with original ASMs and with two recent improved versions.