Neural Collapse (NC) is a recently observed phenomenon in neural networks that characterises the solution space of the final classifier layer when trained until zero training loss. Specifically, NC suggests that the final classifier layer converges to a Simplex Equiangular Tight Frame (ETF), which maximally separates the weights corresponding to each class. By duality, the penultimate layer feature means also converge to the same simplex ETF. Since this simple symmetric structure is optimal, our idea is to utilise this property to improve convergence speed. Specifically, we introduce the notion of \textit{nearest simplex ETF geometry} for the penultimate layer features at any given training iteration, by formulating it as a Riemannian optimisation.

For this explanation let's suppose we were working on a binary classification problem to detect whether or not a transaction is fraudulent. Our model uses characteristics of the user and transaction and returns 1 if the transaction is predicted to be fraudulent and 0 if not. Given that machine learning models are rarely 100% accurate there is going to be a level of risk in deploying this model. If we incorrectly classify a non-fraudulent transaction as fraud then we may well lose that transaction, and possibly even the future customers business. On the other hand, if we incorrectly detect a fraudulent transaction as non-fraudulent then we might stand to lose the value of that transaction. The confusion matrix essentially places the resulting predictions into four groups.

It is common knowledge that machines are better than humans at certain tasks. Need someone to carry out repetitive work on an assembly line? Modern robots, computers and artificially intelligent machines are more than up to the task. Writing, however, is one role that has so far withstood the onslaught of AI. Luckily for some of us, where creativity is required, humans, for the time being at least, still maintain an advantage over machines.

IT solutions provider NIIT Technologies Ltd and AI firm Arago have decided to deepen their partnership to optimise the IT operations of Global In-House Centres (GICs) using artificial intelligence. "As global MNCs in India drive innovation through their GICs, there is a need for them to build deeper partnerships with IT service providers to innovate with intelligent automation," said Arvind Mehrotra, president, Infrastructure Management Services. Umamaheshwar Mudigonda, vice-president, Service Provider Business, of Arago, said that the company had high hopes. "We aim to enable the GICs use cutting-edge AI technologies and implement practical solutions, said a press release.

In this paper, we study instances of complex neural networks, i.e. neural netwo rks with complex topologies. We use Self-Organizing Map neural networks whose n eighbourhood relationships are defined by a complex network, to classify handwr itten digits. We show that topology has a small impact on performance and robus tness to neuron failures, at least at long learning times. Performance may howe ver be increased (by almost 10%) by artificial evolution of the network topo logy. In our experimental conditions, the evolved networks are more random than their parents, but display a more heterogeneous degree distribution.