Contrastive language and vision learning of general fashion concepts.

The steady rise of online shopping goes hand in hand with the development of increasingly complex ML and NLP models. While most use cases are cast as specialized supervised learning problems, we argue that practitioners would greatly benefit from general and transferable representations of products. In this work, we build on recent developments in contrastive learning to train FashionCLIP, a CLIP-like model adapted for the fashion industry. We demonstrate the effectiveness of the representations learned by FashionCLIP with extensive tests across a variety of tasks, datasets and generalization probes. We argue that adaptations of large pre-trained models such as CLIP offer new perspectives in terms of scalability and sustainability for certain types of players in the industry. Finally, we detail the costs and environmental impact of training, and release the model weights and code as open source contribution to the community.

Shopper intent prediction from clickstream e-commerce data with minimal browsing information.

We address the problem of user intent prediction from clickstream data of an e-commerce website via two conceptually different approaches: a hand-crafted feature-based classification and a deep learning-based classification. In both approaches, we deliberately coarse-grain a new clickstream proprietary dataset to produce symbolic trajectories with minimal information. Then, we tackle the problem of trajectory classification of arbitrary length and ultimately, early prediction of limited-length trajectories, both for balanced and unbalanced datasets. Our analysis shows that k-gram statistics with visibility graph motifs produce fast and accurate classifications, highlighting that purchase prediction is reliable even for extremely short observation windows. In the deep learning case, we benchmarked previous state-of-the-art (SOTA) models on the new dataset, and improved classification accuracy over SOTA performances with our proposed LSTM architecture. We conclude with an in-depth error analysis and a careful evaluation of the pros and cons of the two approaches when applied to realistic industry use cases.

Trust and distrust in contradictory information transmission.

We analyse the problem of contradictory information distribution in networks of agents with positive and negative trust. The networks of interest are built by ranked agents with different epistemic attitudes. In this context, positive trust is a property of the communication between agents required when message passing is executed bottom-up in the hierarchy, or as a result of a sceptic agent checking information. These two situations are associated with a confirmation procedure that has an epistemic cost. Negative trust results from refusing verification, either of contradictory information or because of a lazy attitude. We offer first a natural deduction system called SecureNDsim to model these interactions and consider some meta-theoretical properties of its derivations. We then implement it in a NetLogo simulation to test experimentally its formal properties. Our analysis concerns in particular: conditions for consensus-reaching transmissions; epistemic costs induced by confirmation and rejection operations; the influence of ranking of the initially labelled nodes on consensus and costs; complexity results.

Abnormal splicing of ABCA1 pre-mRNA in Tangier disease due to a IVS2 +5G>C mutation in ABCA1 gene.

Two point mutations of ABCA1 gene were found in a patient with Tangier disease (TD): i) G>C in intron 2 (IVS2 +5G>C) and ii) c.844 C>T in exon 9 (R282X). The IVS2 +5G>C mutation was also found in the brother of another deceased TD patient, but not in 78 controls and 33 subjects with low HDL. The IVS2 +5G>C mutation disrupts ABCA1 pre-mRNA splicing in fibroblasts, leading to three abnormal mRNAs: devoid of exon 2 (Ex2-/mRNA), exon 4 (Ex4-/mRNA), or both these exons (Ex2-/Ex4-/mRNA), each containing a translation initiation site. These mRNAs are expected either not to be translated or generate short peptides. To investigate the in vitro effect of IVS2 +5G>C mutation, we constructed two ABCA1 minigenes encompassing Ex1-Ex3 region, one with wild-type (WTgene) and the other with mutant (MTgene) intron 2. These minigenes were transfected into COS1 and NIH3T3, two cell lines with a different ABCA1 gene expression. In COS1 cells, WTgene pre-mRNA was spliced correctly, while the splicing of MTgene pre-mRNA resulted in Ex2-/mRNA. In NIH3T3, no splicing of MTgene pre-mRNA was observed, whereas WTgene pre-mRNA was spliced correctly. These results stress the complexity of ABCA1 pre-mRNA splicing in the presence of splice site mutations.

Reduced collagenolytic activity of matrix metalloproteinases and development of liver fibrosis in the aging rat.

Although moderate fibrosis is a histological hallmark of the aging liver, the molecular mechanisms underlying this phenomenon are little known. Here, we provide a comprehensive description of hepatic collagen expression and metabolism during natural aging in rats. Interstitial collagen accumulated significantly in the oldest animals, mainly in the periportal area (P<0.05, 19- vs. 2-month-old rats). This was ascribed to COL-III protein deposition (P<0.05 vs. 2-month-old rats), rather than COL-I. Conversely, the transcription activity of COL-III gene decreased (P<0.05) during the considered lifespan (2-19-months), whereas COL-I and transforming growth fator-beta1 (TGF-beta1) mRNA content was substantially unchanged. In the aged rats, hepatic matrix metalloproteinases (MMP) activity, (both MMP-1 and MMP-2) dropped significantly (P<0.05), with a concomitant increase of the inactive tissue inhibitor of MMP (TIMP-1)/MMP-1 complex (P<0.05). MMP-2 and TIMP-1 levels were weakly affected. All together, these results suggest that during natural aging, (i) COL III is the protein that accumulates preferentially in the liver; (ii) liver fibrosclerosis is mainly explained by a reduced proteolytic activity of matrix MMP, in which TIMP-1 seems to be a major regulating factor.