On-board processing is emerging as a key enabler for Earth observation (EO) missions, reducing downlink requirements and supporting more autonomous, event-driven operations. Deep convolutional neural networks (CNNs) deliver state-of-the-art performance on many EO tasks, but their memory footprint and computational demands remain challenging for space-qualified hardware. Classical machine learning (CML) pipelines based on hand-crafted spectral and textural features offer a lighter alternative, yet it is unclear how they compare with modern compressed deep models under deployment-relevant efficiency metrics. This work introduces a unified experimental framework that jointly evaluates compressed deep learning (CDL) models and CML ensembles on two representative EO benchmarks: EuroSAT for land-cover classification and HYPERVIEW for hyperspectral soil-property regression. Starting from a common CNN baseline, we evaluate pruning and post-training quantization, and use the selected baseline as the teacher in family-conditioned KD experiments. We contrast the resulting models with optimized tree-based ensembles trained on engineered features. For the benchmark comparisons we measure predictive efficacy, inference time and serialized model size, enabling a systematic comparison of the trade-offs between accuracy, runtime and storage. In our experiments, quantization provides the largest observed storage reduction among the tested CDL variants, moderate pruning can preserve predictive performance more closely in some settings, and the effectiveness of knowledge distillation depends more strongly on the dataset, student design, and distillation setting. Classical ensembles remain attractive when low prediction-stage latency or small serialized models are required and a moderate loss in accuracy is acceptable. For raw-input deployment, their runtime benefit also depends on the cost and implementation of descriptor extraction. The proposed analysis provides empirical guidance for selecting model families and compression strategies when designing future on-board EO systems.

In this paper, we report our experience with “TuringHotel”, a novel extension of the Turing Test based on interactions within mixed communities of Large Language Models (LLMs) and human participants. The classical one-to-one interaction of the Turing Test is reinterpreted in a group setting, where both human and artificial agents engage in time-bounded discussions and, interestingly, are both judges and respondents. This community is instantiated in the novel platform UNaIVERSE (https://unaiverse.io), creating a “World” which defines the roles and interaction dynamics, facilitated by the platform's built-in programming tools. All communication occurs over an authenticated peer-to-peer network, ensuring that no third parties can access the exchange. The platform also provides a unified interface for humans, accessible via both mobile devices and laptops, that was a key component of the experience in this paper. Results of our experimentation involving 17 human participants and 19 LLMs revealed that current models are still sometimes confused as humans. Interestingly, there are several unexpected mistakes, suggesting that human fingerprints are still identifiable but not fully unambiguous, despite the high-quality language skills of artificial participants. We argue that this is the first experiment conducted in such a distributed setting, and that similar initiatives could be of national interest to support ongoing experiments and competitions aimed at monitoring the evolution of large language models over time.

We present a framework for dynamic management of structured parallel processing skeletons on serverless platforms. Our goal is to bring HPC-like performance and resilience to serverless and continuum environments while preserving the programmability benefits of skeletons. As a first step, we focus on the well known Farm pattern and its implementation on the open-source OpenFaaS platform, treating autoscaling of the worker pool as a QoS-aware resource management problem. The framework couples a reusable farm template with a Gymnasium-based monitoring and control layer that exposes queue, timing, and QoS metrics to both reactive and learning-based controllers. We investigate the effectiveness of AI-driven dynamic scaling for managing the farm's degree of parallelism via the scalability of serverless functions on OpenFaaS. In particular, we discuss the autoscaling model and its training, and evaluate two reinforcement learning (RL) policies against a baseline of reactive management derived from a simple farm performance model. Our results show that AI-based management can better accommodate platform-specific limitations than purely model-based performance steering, improving QoS while maintaining efficient resource usage and stable scaling behaviour.

The advent of large-scale pre-trained models has fundamentally transformed the field of artificial intelligence, unlocking new possibilities and achieving unprecedented performance across a wide range of tasks. Yet, these models inherit a fundamental limitation from traditional Machine Learning approaches: their heavy reliance on the i.i.d. assumption, which restricts their capacity to adapt to dynamic, real-world scenarios. We argue that the next major breakthrough in AI lies in developing systems capable of efficient and continual adaptation to evolving environments, where new data and tasks arrive sequentially. This need defines the domain of Continual Learning (CL), a Machine Learning paradigm aimed at developing neural models capable of learning throughout their lifespan without forgetting previous knowledge. In parallel, Parameter-Efficient Fine-Tuning (PEFT) methods have emerged as powerful tools for adapting large models to specific tasks with minimal computational resources. While PEFT techniques can match the performance of full fine-tuning with significantly fewer parameter updates, they remain vulnerable to catastrophic forgetting. This survey bridges the gap between CL and PEFT by focusing on the emerging field of Parameter-Efficient Continual Fine-Tuning (PECFT). We first provide a comprehensive overview of CL strategies and PEFT approaches, then review recent advances in PECFT. We analyze existing methods, compare evaluation protocols and identify open challenges and promising directions for future research. Our aim is to underscore the synergy between CL and PEFT, offering insights and guidance to researchers seeking to build adaptive, scalable and efficient AI systems.

Foundation models have transformed machine learning through large-scale pretraining, massive parameterization, and increased test-time compute. Despite surpassing human performance in several domains, these models remain fundamentally limited in continuous operation, experience accumulation, and personalization, capabilities that are central to adaptive intelligence. While continual learning research has long targeted these goals, its historical focus on in-weight learning, i.e., updating a single model’s parameters to absorb new knowledge, has rendered catastrophic forgetting a persistent challenge. We outline a conceptual framework for modular memory-centric architectures that leverage ICL for rapid adaptation and knowledge accumulation, and IWL for stable updates to model capabilities, thereby mitigating catastrophic forgetting and charting a practical roadmap toward continually learning agents.

Cloud virtual machines are often overprovisioned, creating avoidable cost and operational inefficiency. We present CLOUDADV, an interactive engineer-facing advisory system for cloud instance sizing under workload drift. The system combines zero-shot time-series forecasting with bounded recommendation generation across day-, week-, and month-scale planning horizons. For each query, CLOUDADV constructs a structured decision context from historical utilization, forecast summaries, current VM metadata, candidate instance options, pricing, and explicit sizing heuristics. A higher-capacity LLM is used offline to generate reference recommendations, while a smaller production model is evaluated on the same prompts to assess deployment-time alignment under latency and cost constraints. Evaluation prioritizes downstream recommendation quality using simulated Azure cost savings and ex-post exceedance, with rolling-origin forecast accuracy reported as a secondary diagnostic against classical and supervised baselines. In a case study of seven production VMs, the reference recommendations reduce simulated monthly cost from about $1,503 to $708, yielding $795/month in savings (52.9

AI model hubs provide access to a rapidly growing collection of powerful pre-trained models, enabling off-the-shelf mixture-of-experts systems with different routing strategies. However, this rapid growth poses two fundamental challenges: scaling model selection across thousands of experts and continually updating routing mechanisms as new models and tasks are introduced. In this paper, we formalise this setting as Continual Model Routing (CMR) and propose , a new large-scale benchmark simulating realistic hub expansion and including over 2,000 candidate models. Finally, we introduce , a contrastive embedding approach for efficient continual model routing via domain-stratified coreset replay and checkpoint-based anchoring. Extensive empirical results and ablations show that CARvE significantly outperforms zero-shot retrieval, fine-tuning, and adapter-merging baselines in model, family, and domain-level accuracy.

Compositionality is one of the fundamental abilities of the human reasoning process, that allows to decompose a complex problem into simpler elements. Such property is crucial also for neural networks, especially when aiming for a more efficient and sustainable AI framework. We propose a compositional approach by ensembling zero-shot a set of expert models, assessing our methodology using a challenging benchmark, designed to test compositionality capabilities. We show that our Expert Composition method is able to achieve a much higher accuracy than baseline algorithms while requiring less computational resources, hence being more efficient.

Continual learning–the ability to acquire, retain, and refine knowledge over time–has always been fundamental to intelligence, both human and artificial. Historically, different AI paradigms have acknowledged this need, albeit with varying priorities: early expert and production systems focused on incremental knowledge consolidation, while reinforcement learning emphasised dynamic adaptation. With the rise of deep learning, deep continual learning has primarily focused on learning robust and reusable representations over time to solve sequences of increasingly complex tasks. However, the emergence of Large Language Models (LLMs) and foundation models has raised the question: Do we still need continual learning when centralised, monolithic models can tackle diverse tasks with access to internet-scale knowledge? We argue that continual learning remains essential for three key reasons: (i) continual pre-training is still necessary to ensure foundation models remain up to date, mitigating knowledge staleness and distribution shifts while integrating new information; (ii) continual fine-tuning enables models to specialise and personalise, adapting to domain-specific tasks, user preferences, and real-world constraints without full retraining, avoiding the need for computationally expensive long context-windows; (iii) continual compositionality offers a scalable and modular approach to intelligence, enabling the orchestration of foundation models and agents to be dynamically composed, recombined, and adapted. While continual pre-training and fine-tuning are explored as niche research directions, we argue it is continual compositionality that will mark the rebirth of continual learning. The future of AI will not be defined by a single static model but by an ecosystem of continually evolving and interacting models, making continual learning more relevant than ever.

In today’s digital landscape, online social networks (OSNs) facilitate rapid information dissemination. However, they also serve as conduits for misinformation, leading to severe real-world consequences such as public panic, social unrest, and the erosion of institutional trust. Existing rumor influence minimization strategies predominantly rely on static models or specific diffusion mechanisms, restricting their ability to dynamically adapt to the evolving nature of misinformation. To address this gap, this paper proposes a novel misinformation influence mitigation framework that integrates Graph Neural Networks (GNNs) with continual learning and employs a Node Blocking strategy as its intervention approach. The framework comprises three key components: (1) a Dataset Generator, (2) a GNN Model Trainer, and (3) an Influential Node Identifier. Given the scarcity of real-world data on misinformation propagation, the first component simulates misinformation diffusion processes within social networks, leveraging the Human Individual and Social Behavior (HISB) model as a case study. The second component employs GNNs to learn from these synthetic datasets and predict the most influential nodes susceptible to misinformation. Subsequently, these nodes are strategically targeted and blocked to minimize further misinformation spread. Finally, the continual learning mechanism ensures the model dynamically adapts to evolving network structures and propagation patterns. Beyond evaluating the Human Individual and Social Behavior (HISB) propagation model, we empirically demonstrate that our framework is propagation-model agnostic by reproducing the pipeline under Independent Cascade and Linear Threshold with consistent gains over baselines. Finally, we introduce a truth-aware intervention rule that gates and weights actions by an external veracity score at detection time, selecting most influential nodes. This addition ensures interventions are enacted only when content is likely false, aligning the method with responsible deployment. Experimental evaluations conducted on multiple benchmark datasets demonstrate the superiority of the proposed node blocking framework over state-of-the-art methods. Our results indicate a statistically significant reduction in misinformation spread, with non-parametric statistical tests yielding p-values below 0.001 (p<0.001), confirming the robustness of our approach. This work presents a scalable and adaptable solution for misinformation containment, contributing to the development of more reliable and trustworthy online information ecosystems.

The birth of Foundation Models brought unprecedented results in a wide range of tasks, from language to vision, to robotic control. These models are able to process huge quantities of data, and can extract and develop rich representations, which can be employed across different domains and modalities. However, they still have issues in adapting to dynamic, real-world scenarios without retraining the entire model from scratch. In this work, we propose the application of Continual Learning and Compositionality principles to foster the development of more flexible, efficient and smart AI solutions.

Continually learning from non-independent and identically distributed (non-i.i.d.) data poses a significant challenge in deep learning, particularly in resource-constrained environments. Visual models trained via supervised learning often suffer from overfitting, catastrophic forgetting, and biased representations when faced with sequential tasks. In contrast, pre-trained language models demonstrate greater robustness in managing task sequences due to their generalized knowledge representations, albeit at the cost of high computational resources. Leveraging this advantage, we propose a novel learning strategy, Continual Visual Mapping (CVM), which continuously maps visual representations into a fixed knowledge space derived from a language model. By anchoring learning to this fixed space, CVM enables training small, efficient visual models, making it particularly suited for scenarios where adapting large pre-trained visual models is computationally or data-prohibitive. Empirical evaluations across five benchmarks demonstrate that CVM consistently outperforms state-of-the-art continual learning methods, showcasing its potential to enhance generalization and mitigate challenges in resource-constrained continual learning settings.

Continual learning (CL) is the sub-field of machine learning concerned with accumulating knowledge in dynamic environments. So far, CL research has mainly focused on incremental classification tasks, where models learn to classify new categories while retaining knowledge of previously learned ones. Here, we argue that maintaining such a focus limits both theoretical development and practical applicability of CL methods. Through a detailed analysis of concrete examples - including multi-target classification, robotics with constrained output spaces, learning in continuous task domains, and higher-level concept memorization - we demonstrate how current CL approaches often fail when applied beyond standard classification. We identify three fundamental challenges: (C1) the nature of continuity in learning problems, (C2) the choice of appropriate spaces and metrics for measuring similarity, and (C3) the role of learning objectives beyond classification. For each challenge, we provide specific recommendations to help move the field forward, including formalizing temporal dynamics through distribution processes, developing principled approaches for continuous task spaces, and incorporating density estimation and generative objectives. In so doing, this position paper aims to broaden the scope of CL research while strengthening its theoretical foundations, making it more applicable to real-world problems.

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

We introduce the concept of collaborative awareness as a means to enhance interoperability, resilience and self regulation in synthetic agent collectives. We discuss the theoretical, computational and engineering framework of collaborative awareness built by the EU project EMERGE, and its application to distributed robotic systems.

In the Cloud-Edge Continuum, dynamic infrastructure change and variable workloads complicate efficient resource management. Centralized methods can struggle to adapt, whilst purely decentralized policies lack global oversight. This paper proposes a hybrid framework using Graph Neural Network (GNN) embeddings and collaborative multi-agent reinforcement learning (MARL). Local agents handle neighbourhood-level decisions, and a global orchestrator coordinates system-wide. This work contributes to decentralized application placement strategies with centralized oversight, GNN integration and collaborative MARL for efficient, adaptive and scalable resource management.

The ability to learn continuously over time remains a major challenge for modern machine learning systems, even in the era of Foundation Models. While the rich representations learned by large pre-trained models can partially mitigate catastrophic forgetting, they still struggle to adapt efficiently to evolving data distributions. A key challenge remains, how to continually add new knowledge to a large pretrained model in a way that is scalable and computationally efficient over long task sequences. In this work, we introduce GLAM, a simple and effective framework for class-incremental continual learning based on LoRA adapters merging. For each task, GLAM trains a lightweight low-rank adapter with an importance scalar, incurring minimal computational overhead. Adapters are then pruned, rescaled, and sequentially grouped to enable structured knowledge reuse and limit parameter growth. At inference, all groups are combined into a single module, ensuring constant computational cost regardless of the number of tasks. We evaluate GLAM on vision benchmarks with sequences of up to 50 tasks, significantly extending beyond standard protocols. GLAM achieves the highest accuracy across the evaluated benchmarks. Compared with the baseline attaining the highest average accuracy across benchmarks, it uses 16–19% of the trainable parameters and reduces training time by approximately 63–74%, demonstrating efficient and scalable continual learning. Our source code is publicly available at https://github.com/atlas-luiss/GLAM.

The recent focus and release of pre-trained models have been a key components to several advancements in many fields (e.g. Natural Language Processing and Computer Vision), as a matter of fact, pre-trained models learn disparate latent embeddings sharing insightful representations. On the other hand, Reinforcement Learning (RL) focuses on maximizing the cumulative reward obtained via agent's interaction with the environment. RL agents do not have any prior knowledge about the world, and they either learn from scratch an end-to-end mapping between the observation and action spaces or, in more recent works, are paired with monolithic and computationally expensive Foundational Models. How to effectively combine and leverage the hidden information of different pre-trained models simultaneously in RL is still an open and understudied question. In this work, we propose Weight Sharing Attention (WSA), a new architecture to combine embeddings of multiple pre-trained models to shape an enriched state representation, balancing the tradeoff between efficiency and performance. We run an extensive comparison between several combination modes showing that WSA obtains comparable performance on multiple Atari games compared to end-to-end models. Furthermore, we study the generalization capabilities of this approach and analyze how scaling the number of models influences agents' performance during and after training.

Time series and sequential data are widespread in many real-world environments. However, implementing physical and adaptive dynamical systems remains a challenge. Direct Feedback Alignment (DFA) is a learning algorithm for neural networks that overcomes some of the limits of backpropagation and can be implemented in neuromorphic hardware (e.g., photonic accelerators). Until now, DFA has been investigated mainly for feedforward architectures. We adapt DFA for both “vanilla” and gated recurrent networks. Unlike backpropagation, the update rule of our DFA can be applied in parallel across time steps, thus removing the sequential propagation of errors. We benchmark DFA on 4 datasets for sequence classification tasks. Although backpropagation still achieves a better predictive accuracy, our DFA shows promising results, especially for environments and physical systems where backpropagation is unavailable.