187,565 research outputs found
Training Generative Adversarial Networks via Stochastic Nash Games
No full textGenerative adversarial networks (GANs) are a class of generative models with two antagonistic neural networks: a generator and a discriminator. These two neural networks compete against each other through an adversarial process that can be modeled as a stochastic Nash equilibrium problem. Since the associated training process is challenging, it is fundamental to design reliable algorithms to compute an equilibrium. In this article, we propose a stochastic relaxed forward-backward (SRFB) algorithm for GANs, and we show convergence to an exact solution when an increasing number of data is available. We also show convergence of an averaged variant of the SRFB algorithm to a neighborhood of the solution when only a few samples are available. In both cases, convergence is guaranteed when the pseudogradient mapping of the game is monotone. This assumption is among the weakest known in the literature. Moreover, we apply our algorithm to the image generation problem.</p
Composite control Lyapunov functions for robust stabilization of constrained uncertain dynamical systems
No full textThis work presents innovative scientific results on the robust stabilization of constrained uncertain dynamical systems via Lyapunov-based state feedback control.
Given two control Lyapunov functions, a novel class of smooth composite control Lyapunov functions is presented. This class, which is based on the R-functions theory, is universal for the stabilizability of linear differential inclusions and has the following property. Once a desired controlled invariant set is fixed, the shape of the inner level sets can be made arbitrary close to any given ones, in a smooth and non-homothetic way. This procedure is an example of ``merging'' two control Lyapunov functions.
In general, a merging function consists in a control Lyapunov function whose gradient is a continuous combination of the gradients of the two parents control Lyapunov functions.
The problem of merging two control Lyapunov functions, for instance a global control Lyapunov function with a large controlled domain of attraction and a local one with a guaranteed local performance, is considered important for several control applications. The main reason is that when simultaneously concerning constraints, robustness and optimality, a single Lyapunov function is usually suitable for just one of these goals, but ineffective for the others.
For nonlinear control-affine systems, both equations and inclusions, some equivalence properties are shown between the control-sharing property, namely the existence of a single control law which makes simultaneously negative the Lyapunov derivatives of the two given control Lyapunov functions, and the existence of merging control Lyapunov functions.
Even for linear systems, the control-sharing property does not always hold, with the remarkable exception of planar systems.
For the class of linear differential inclusions, linear programs and linear matrix inequalities conditions are given for the the control-sharing property to hold.
The proposed Lyapunov-based control laws are illustrated and simulated on benchmark case studies, with positive numerical results
Efficacy of new drugs as induction treatment before single or tandem autologous stem cell transplantation in newly diagnosed multiple myeloma patients: a single centre experience
No full textThe outcome of patients affected by Multiple Myeloma (MM) has markedly improved over the past decade, both in young and elderly patients. In the past years, conventional therapy, such as melphalan plus prednisone (MP), was the only active treatment against MM. Since the 1980s, high doses of chemotherapy plus autologous stem cell transplantation (ASCT) proved to be the most suitable option for young newly diagnosed multiple myeloma (YNDMM) patients. More recently, new active classes of drugs, such as proteasome inhibitors-PIs (e.g., bortezomib) and immunomodulatory drugs-IMiDs (e.g., thalidomide and lenalidomide) became the standard of care, alone or in association with the old agents, either for fırst-line therapy in the transplant and non-transplant settings, or for the treatment of relapsed disease.
Actually, three-drugs combinations, including at least bortezomib or lenalidomide in combination with dexamethasone, followed by ASCT is considered the optimal choice for all YNDMM.
Patients and Methods: we retrospectively analyzed our Centre experience in the last 25 years comparing the results obtained with old drugs- that is conventional chemotherapy- versus novel agents –such as lenalidomide and bortezomib- in YNDMM. We also performed an additional analysis on the effect of a single versus tandem ASCT either with old agents or with novel agents.
Results: Between August 1989 and May 2014, 258 YNDMM patients underwent ASCT. The median age was 54 years (range, 18-69), 137 were men. As induction treatment, between October 1988 and October 2008, 173/258 patients received old drugs, i.e. vincristine, doxorubicin and dexamethasone (VAD; n=167) or MP (n=6), while 85/258 patients, between February 2005 and November 2013, were treated with novel agents, i.e. velcade-based (n=67) or IMiD-based regimens (n=18). All 258 patients received high doses of melphalan and single (n=153) or tandem (n=105) ASCT. Overall, after induction, 67 patients (25.9%) achieved complete response (CR), near CR (nCR) or very good partial response (VGPR). More in detail, among patients treated with new drugs, a CR/nCR/VGPR was observed in 36/85 patients (42.3%) after induction, in 36/85 patients (42.3%) after single ASCT and in 27/50 patients (54%) after tandem ASCT. No differences were observed in terms of response and survival between IMiDs or bortezomib-based regimens. For patients treated with old drugs, a CR/nCR/VGPR was recorded in 31/173 (17.9%) after induction, in 50/173 patients (28.9%) after single ASCT and in 19/55 patients (34.5%) after tandem ASCT.
Overall survival (OS) and progression-free survival (PFS) at 10 years, for all 258 patients, was 44.4% and 22.5%, respectively. OS and PFS were better for patients in CR/nCR/VGPR after induction compared to those in partial response or stable disease (OS: 62.1% vs 40.7%, p=0.06 and PFS: 36.2% vs 17.2%, p=0.06). In addition, OS was slightly better for patients treated with new drugs than those treated with old drugs. In fact, OS at 8 years was 66.1% (IC 95%:53.5-81.7) for the first group vs 51.5% for the second group, respectively (p = 0.26 n.s.). PFS was significantly improved for patients treated with new drugs than those treated with old drugs. In fact, PFS at 8 years was 55% for the first group vs 25.3% for the second group, respectively (p = 0.0047).
Supposing that the impact on OS was influenced by the salvage treatments used after the progression of disease, we further analyzed patients that relapsed. Among our 258 patients, 144 presented a first relapse.
Our cohort of relapsed patients was divided in 4 different groups:
1) 51 patients treated with old agents as first and second line therapy (35.4%);
2) 79 patients treated with old agents for first line therapy and novel agents for second line therapy (54.8%);
3) 2 patients treated with novel agents in induction therapy and subsequent old agents in second line therapy for worsening clinical condition (1.4%);
4) 12 patients treated with novel agents both in first and second line treatment (8.3%).
Our analysis was focused on group 1 and 2. The group 3 was not considered for the small number of patients and because the choice of treatment was based exclusive on clinical worsening condition; the group 4 was not included for the small number of patients and because follow up is too short.
OS at 10 years for patients of group 2 was significantly higher than patients of group 1 (20.4 vs 2.4%; p < 0.0001). Also PFS at 10 years showed better results for patients of group 2 (10% vs 2.3%; p = 0.02). The PFS2 at 10 years, considered as the interval from the start of the first line treatment to progression after second-line therapy or death from any cause, showed better results for patients of group 2 than patients of group 1 (25.7% vs 9.2%; p = 0.0002).
Conclusions: Our experience confirmed that novel agents provide better outcome and deeper responses as induction treatment for YNDMM patients and tandem ASCT still improve the responses in these patients. Moreover, novel agents have also a significant impact in the subsequent lines of treatments, showing better results in terms of PFS2. In the light of the newer IMiDs and PIs, the challenge is to assess the exact role of ASCT in the modern setting; our next effort will be the evaluation of YNDMM patients treated with the new-generation agents (carfilzomib and others) in terms of outcome and safety, and the role of ASCT even in this setting, in a larger cohort of patients and with a appropriate follow up
Albertus Linnenkugel, De Lucillo Tarrhaeo epigrammatum poeta, grammatico, rhetore (Rhetorische Studien, fasc. 13)
No full textBoulanger André. Albertus Linnenkugel, De Lucillo Tarrhaeo epigrammatum poeta, grammatico, rhetore (Rhetorische Studien, fasc. 13). In: Revue des Études Anciennes. Tome 29, 1927, n°1. p. 103
- …
