The publication record.
A small lab that publishes with universities rather than instead of them. The work runs from thalamocortical circuitry to seismic facies to the reporting standards that medical AI is now reviewed against.
Artificial intelligence in dental research: Checklist for authors, reviewers, readers
Written in the ITU/WHO Focus Group on AI for Health, topic group on dental diagnostics. The lab's most-cited work by a wide margin.
Seismic Facies Analysis: A Deep Domain Adaptation Approach
EarthAdaptNet. Trained on the F3 block offshore Netherlands, tested on Penobscot, Canada.
From Knowing to Doing: A Principled Architecture for Procedurally Competent Agents
The HOMINIS architecture paper. Led by the PICUS Lab at University of Naples Federico II, with Real AI as industrial partner.
A century of well logs, drawn on paper.
Before logging went digital, the tools wrote to paper. Those sheets were later scanned, and the scans are what an operator has: an image of a curve, not the curve. VeerNet classifies the curves in a raster log and turns them back into numbers.
FIGURES FROM NASIM, PATWARDHAN, MAITI, MARRONE AND SINGH, VEERNET, JOURNAL OF IMAGING 9(7):136, 2023, DOI 10.3390/JIMAGING9070136. PUBLISHED OPEN ACCESS UNDER CC BY 4.0 AND REPRODUCED UNDER THAT LICENCE.
What a flow model looks like from the inside.
Flow-based models are exactly invertible, so the same network runs forwards to sample and backwards to score. That symmetry is the whole design, and it is easier to see than to describe.
BOTH FIGURES ARE THE LAB'S OWN, PUBLISHED WITH THE PAPER IN 2019. THE WORK REACHED STATE OF THE ART AMONG FLOW-BASED MODELS ON IMAGENET 32×32 AND 64×64, AND CONVERGED AROUND TEN TIMES FASTER THAN GLOW.
The papers carry names, not a logo.
A neuroscientist in Berlin, a geophysicist in Calgary, a generative-models researcher in Chongqing, a machine-learning researcher who moved from Michigan to Naples. The lab has always been a handful of people in different countries who publish together, several of them under a university affiliation as well as ours.
Tarry SinghFOUNDERAssen, Netherlands
Jie MeiCOMPUTATIONAL NEUROSCIENCECharité Berlin, then Toronto
Tannistha MaitiGEOPHYSICSUniversity of Calgary, Toronto
Narendra PatwardhanMACHINE LEARNINGMichigan Tech, then Naples
Huadong LiaoGENERATIVE MODELSChongqing, China
The code behind the papers, and not much else.
Three repositories carry original work, one of them released by its author on his own account. The rest of the organisation is forks and teaching material, which we would rather not dress up as research.
