References of "Speeches/Talks"
     in
Bookmark and Share    
Full Text
See detailLeveraging state-of-the-art architectures by enriching training information - a case study
Sölter, Jan; Proverbio, Daniele; Baniasadi, Mehri et al

Speeches/Talks (2021)

Our working hypothesis is that key factors in COVID-19 imaging are the available imaging data and their label noise and confounders, rather than network architectures per se. Thus, we applied existing ... [more ▼]

Our working hypothesis is that key factors in COVID-19 imaging are the available imaging data and their label noise and confounders, rather than network architectures per se. Thus, we applied existing state-of-the-art convolution neural network frameworks based on the U-Net architecture, namely nnU-Net [3], and focused on leveraging the available training data. We did not apply any pre-training nor modi ed the network architecture. First, we enriched training information by generating two additional labels for lung and body area. Lung labels were created with a public available lung segmentation network and weak body labels were generated by thresholding. Subsequently, we trained three di erent multi-class networks: 2-label (original background and lesion labels), 3-label (additional lung label) and 4-label (additional lung and body label). The 3-label obtained the best single network performance in internal cross-validation (Dice-Score 0.756) and on the leaderboard (Dice- Score 0.755, Haussdor 95-Score 57.5). To improve robustness, we created a weighted ensemble of all three models, with calibrated weights to optimise the ranking in Dice-Score. This ensemble achieved a slight performance gain in internal cross-validation (Dice-Score 0.760). On the validation set leaderboard, it improved our Dice-Score to 0.768 and Haussdor 95- Score to 54.8. It ranked 3rd in phase I according to mean Dice-Score. Adding unlabelled data from the public TCIA dataset in a student-teacher manner signi cantly improved our internal validation score (Dice-Score of 0.770). However, we noticed partial overlap between our additional training data (although not human-labelled) and  nal test data and therefore submitted the ensemble without additional data, to yield realistic assessments. [less ▲]

Detailed reference viewed: 34 (4 UL)
Full Text
See detailEventually everything connects
Lallemand, Carine UL

Speeches/Talks (2018)

If you are reading these lines, you are probably curious about the type of experience that you would have as an attendee of this talk. You might wonder whether the topic would match your interests ... [more ▼]

If you are reading these lines, you are probably curious about the type of experience that you would have as an attendee of this talk. You might wonder whether the topic would match your interests, whether the speaker will be good enough to satisfy your expectations, whether you will feel inspired, exhilarated, or whether you will have concrete tools to bring back to your work. While you are usually the ones shaping people’s experiences, you are striving for nice experiences as well. Designing for human experiences is one of the most challenging yet fascinating activities. It is a responsibility that we should embrace with humility and dedication. To face the complexity of our mission, we need to draw on theoretical knowledge, methodological skills and of course on our shared professional expertise, as a community. While UX practitioners are working hard at the front lines to design better products or services, scientists work in the shadows to develop and consolidate a myriad of novel and highly valuable UX methods. During this talk, you will discover the ever-growing UX toolbox that could greatly support you in collecting richer, insightful and more valid data. From scientific theories to pragmatic methods, from academia to industry, from Luxembourg to Puerto Rico… Eventually everything connects. [less ▲]

Detailed reference viewed: 96 (1 UL)
Full Text
See detailMulti-scale modelling of fracture
Bordas, Stéphane UL; Kerfriden, Pierre; Beex, Lars et al

Speeches/Talks (2016)

We present recent models on complexity reduction for computational fracture mechanics

Detailed reference viewed: 192 (7 UL)
Full Text
See detailSimulating topological changes in real time for surgical assistance
Bordas, Stéphane UL; Kerfriden, Pierre; Courtecuisse, Hadrien et al

Speeches/Talks (2016)

Detailed reference viewed: 449 (38 UL)
Full Text
See detailMulti-scale methods for fracture: model learning across scales, digital twinning and factors of safety
: primer on Bayesian Inference
Bordas, Stéphane UL; Hale, Jack UL; Beex, Lars UL et al

Speeches/Talks (2015)

Fracture and material instabilities originate at spatial scales much smaller than that of the structure of interest: delamination, debonding, fibre break- age, cell-wall buckling, are examples of nano ... [more ▼]

Fracture and material instabilities originate at spatial scales much smaller than that of the structure of interest: delamination, debonding, fibre break- age, cell-wall buckling, are examples of nano/micro or meso-scale mechanisms which can lead to global failure of the material and structure. Such mech- anisms cannot, for computational and practical reasons, be accounted at structural scale, so that acceleration methods are necessary. We review in this presentation recently proposed approaches to reduce the computational expense associated with multi-scale modelling of frac- ture. In light of two particular examples, we show connections between algebraic reduction (model order reduction and quasi-continuum methods) and homogenisation-based reduction. We open the discussion towards suitable approaches for machine-learning and Bayesian statistical based multi-scale model selection. Such approaches could fuel a digital-twin concept enabling models to learn from real-time data acquired during the life of the structure, accounting for “real” environmental conditions during predictions, and, eventually, moving beyond the era of factors of safety. [less ▲]

Detailed reference viewed: 196 (5 UL)