Link prediction is an unbalanced early retrieval problem, whose goal is to prioritize a small cohort of positive links on top of a list largely populated by unlabelled links. Differently from binary classification, here the evaluation focuses on how the predictor prioritizes the positive class because, in practice, a negative class does not exist. Previous studies explained that AUC-ROC is not apt for unbalanced class problems and is misleading for early retrieval problems, therefore standard AUC-ROC is not appropriate for evaluation of link prediction. However, some scholars argue that an AUC-ROC like evaluation accounting for the relative positioning of the few positive links among the vastness of unlabelled links remains a valid concept to pursue. Here we propose the area under the magnified ROC (AUC-mROC), a new measure that adjusts the standard AUC-ROC to work also for unbalanced early retrieval problems such as link prediction.

Link prediction is an iconic problem in complex networks because deals with the ability to predict nonobserved existing or future parts of the network structure. The impact of this prediction on real applications can be disruptive: from prediction of covert links between terrorists in their social networks to repositioning of drugs in molecular diseasome networks. Here we compare: (1) an ensemble meta-learning method (Ghasemian et al.), which uses an artificial intelligence (AI) stacking strategy to create a single meta-model from hundreds of other models; (2) a structural predictability method (SPM, Lü et al.), which relies on a theory derived from quantum mechanics and does not assume any model; (3) a modelling rule named Cannistraci-Hebb (CH, Muscoloni et al.), which relies on one brain-bioinspired model adapting to the intrinsic network structure.We conclude that brute-force stacking of algorithms by AI does not perform better than (and is often significantly outperformed by) SPM and one simple brain-bioinspired rule such as CH. This agrees with the Gödel incompleteness: stacking is optimal but incomplete, you cannot squeeze out more than what is already in your features. Hence, we should also pursue AI that resembles human-like physical ‘understanding’ of simple generalized rules associated to complexity. The future might be populated by AI that ‘steals for us the fire from Gods’, towards machine intelligence that creates new rules rather than stacking the ones already known.

Sparse training (ST) aims to ameliorate deep learning by replacing fully connected artificial neural networks (ANNs) with sparse or ultra-sparse ones, such as brain networks are, therefore it might benefit to borrow brain-inspired learning paradigms from complex network intelligence theory. Here, we launch the ultrasparse advantage challenge, whose goal is to offer evidence on the extent to which ultra-sparse (around 1% connection retained) topologies can achieve any leaning advantage on fully connected. Epitopological learning is a field of network science and complex network intelligence that studies how to implement learning on complex networks by changing the shape of their connectivity structure (epitopological plasticity). One way to implement Epitopological (epi- means new) Learning is via link prediction: predicting the likelihood of nonobserved links to appear in the network. Cannistraci-Hebb learning theory inspired the CH3-L3 network automata rule for link prediction which is effective for general-purpose link prediction. Here, starting from CH3-L3 we propose Epitopological Sparse Meta-deep Learning (ESML) to apply Epitopological Learning to sparse training. In empirical experiments, we find that ESML learns ANNs with ultra-sparse hyperbolic (epi-)topology in which emerges a community layer organization that is meta-deep (meaning that each layer also has an internal depth due to powerlaw node hierarchy). Furthermore, we discover that ESML can in many cases automatically sparse the neurons during training (arriving even to 30% neurons left in hidden layers), this process of node dynamic removal is called percolation. Starting from this network science evidence, we design Cannistraci-Hebb training (CHT), a 4-step training methodology that put ESML at its heart. We conduct experiments on 6 datasets and 3 network structures (MLPs, VGG16, ResNet50) comparing CHT to dynamic sparse training SOTA algorithms and fully connected network. The results indicate that, with a mere 1% of links retained during training, CHT surpasses fully connected networks on VGG16 and ResNet50. This key finding is an evidence for ultra-sparse advantage and signs a milestone in deep learning. CHT acts akin to a gradient-free oracle which adopts CH3-L3 based epitopological learning to guide the placement of new links in the ultra-sparse network topology to facilitate sparse-weight gradient learning, and this in turn reduces the convergence time of ultra-sparse training. Finally, CHT offers first examples of parsimony dynamic sparse training because, in many datasets, it canretain network performance by percolating and significantly reducing the node network size.

Many complex networks have a connectivity that might be only partially detected or that tends to grow over time, hence the prediction of non-observed links is a fundamental problem in network science. The aim of topological link prediction is to forecast these non-observed links by only exploiting features intrinsic to the network topology. It has a wide range of real applications, like suggesting friendships in social networks or predicting interactions in biological networks.The Cannistraci-Hebb theory is a recent achievement in network science that includes a theoretical framework to understand local-based link prediction on paths of length n. In this study we introduce two innovations: theory of modelling (science) and theory of realization (engineering). For the theory of modelling we first recall a definition of network automata as a general framework for modelling the growth of connectivity in complex networks. We then show that several deterministic models previously developed fall within this framework and we introduce novel network automata following the Cannistraci-Hebb rule. For the theory of realization, we present how to build adaptive network automata for link prediction, which incorporate multiple deterministic models of self-organization and automatically choose the rule that better explains the patterns of connectivity in the network under investigation. We compare Cannistraci-Hebb adaptive (CHA) network automaton against state-of-the-art link prediction methods such as structural perturbation method (SPM), stochastic block models (SBM) and artificial intelligence algorithms for graph embedding. CHA displays an overall higher link prediction performance across different evaluation frameworks on 1386 networks. Finally, we highlight that CHA offers the key advantage to explicitly explain the mechanistic rule of self-organization which leads to the link prediction performance, whereas SPM and graph embedding not. In comparison to CHA, SBM unfortunately shows irrelevant and unsatisfactory performance demonstrating that SBM modelling is not adequate for link prediction in real networks.

Many real complex systems present multilayer structure where high-order metadata on one layer refers to dyadic data on a lower layer. Significant progresses to analyse high-order metadata under the assumption of community organization have been done. However, there are no planted communities in real-world networks, and the necessity of new frameworks to analyze high-order metadata regardless of community organization has been raised.Here, we propose to adopt hyperedge organization. Predicting ‘entanglements’ between a hyperedge and nodes scattered in the rest of the network might suggest structural or functional liaisons, without assumption of any community organization. We introduce a novel concept: hyperedge entanglement (HE), which associates to each hyperedge an entangled hyperedge, by means of a network operator that predicts significant ‘interactions at distance’ between network nodes and existing hyperedges. We also introduce a new challenge termed hyperedge entanglement prediction (HEP), and an algorithm to perform this task. We evaluated HEP performance on social, biological and synthetic data where, given only topology and hyperedges (such as communities or functional modules), the goal is to predict whether nodes not connected to a certain hyperedge might be candidates for a significant entanglement. Finally, as real application in diseasome systems biomedicine, we perform HEP on the human protein interactome to predict unknown gene entanglements with the COPD disease gene hyperedge. HEP predictions are validated by biological experiments, enlarging our understanding of molecular mechanisms behind COPD/aneurysm comorbidity.