Artificial intelligence: Towards a better understanding of the underlying mechanisms

#artificialintelligence

The automatic identification of complex features in images has already become a reality thanks to artificial neural networks. Some examples of software exploiting this technique are Facebook's automatic tagging system, Google's image search engine and the animal and plant recognition system used by iNaturalist. We know that these networks are inspired by the human brain, but their working mechanism is still mysterious. New research, conducted by SISSA in association with the Technical University of Munich and published for the 33rd Annual NeurIPS Conference, proposes a new approach for studying deep neural networks and sheds new light on the image elaboration processes that these networks are able to carry out. Similar to what happens in the visual system, neural networks used for automatic image recognition analyse the content progressively, through a chain of processing stages.


Artificial intelligence: Towards a better understanding of the underlying mechanisms

#artificialintelligence

The automatic identification of complex features in images has already become a reality thanks to artificial neural networks. Some examples of software exploiting this technique are Facebook's automatic tagging system, Google's image search engine and the animal and plant recognition system used by iNaturalist. We know that these networks are inspired by the human brain, but their working mechanism is still mysterious. New research, conducted by SISSA in association with the Technical University of Munich and published for the 33rd Annual NeurIPS Conference, proposes a new approach for studying deep neural networks and sheds new light on the image elaboration processes that these networks are able to carry out. Similar to what happens in the visual system, neural networks used for automatic image recognition analyse the content progressively, through a chain of processing stages.


Intrinsic dimension of data representations in deep neural networks

arXiv.org Machine Learning

Deep neural networks progressively transform their inputs across multiple processing layers. What are the geometrical properties of the representations learned by these networks? Here we study the intrinsic dimensionality (ID) of data-representations, i.e. the minimal number of parameters needed to describe a representation. We find that, in a trained network, the ID is orders of magnitude smaller than the number of units in each layer. Across layers, the ID first increases and then progressively decreases in the final layers. Remarkably, the ID of the last hidden layer predicts classification accuracy on the test set. These results can neither be found by linear dimensionality estimates (e.g., with principal component analysis), nor in representations that had been artificially linearized. They are neither found in untrained networks, nor in networks that are trained on randomized labels. This suggests that neural networks that can generalize are those that transform the data into low-dimensional, but not necessarily flat manifolds.


Scientists hit back at dark matter skeptics who say it is is a myth

Daily Mail - Science & tech

Researchers say they've dispelled skeptics of the most abundant, mysterious, and not to mention, hypothetical, substances in the universe: dark matter. In a recent paper, researchers from the International School for Advanced Studies (SISSA) say that they've disproved alternate theories that seem to preclude the existence of dark matter by studying one of the most puzzling questions in astrophysics. The phenomena, known galaxy rotation curves, has to do with discrepancies with the way objects act within a galaxy. While the laws of astrophysics dictate that solar systems and objects that rotate around the outskirts of spiral galaxies should be moving at slower speeds due to lower amounts of luminous matter, the observed velocity in spiral galaxies like our own, the Milky Way, are uniform. In order to reconcile that observation with our understanding of astrophysics as we know the, specifically Kepler's Laws, scientists have hypothesized that there is matter on the outskirts, and indeed everywhere in our galaxy that we're not seeing -- dark matter.


Strange 'Zoo' Of Dusty Discs Captured Around Newly-Formed Stars 230-550 Light-Years Away

International Business Times

The European Space Observatory (ESO) shared Wednesday an incredible image showcasing weird discs of dust around a bunch of newly-formed stars sitting in the cosmos. The image, taken by the observatory's Very Large Telescope (VLT) in Chile, focuses on the backyard of stars belonging to a class called T Tauri. The members of this group shine with different intensity in the night sky and are just 10 million years old, which is relatively young by cosmic standards. Due to the dazzling glow of these bodies, regions around them often become hard to observe. Plus, the distance to the objects -- approximately 230 to 550 light years from Earth -- further adds to the problem, making it even more difficult for astronomers to see what's present in their proximity.