186698 research outputs found
Sort by
The Stochastic sub-Gradient Method under Heavy-Tailed Noise
No full textThis thesis is dedicated to the study of the Stochastic sub-Gradient Method (SsGM) applied to data that are affected by heavy-tailed noise. The emergence of heavy-tailed noise in modern datasets has been investigated intensively in recent years. This type of noise affects negatively the performances of most optimization and machine learning methods, which instead are conceived with well-behaved light-tailed noise in mind. As SsGM is arguably the most adopted method in modern machine learning applications, it is important to characterize its robustness to noise. To increase the robustness of the method, clipping - i.e. the projection of the sub-gradients in a certain ball before updating the iterates - has emerged as one of the simplest yet effective techniques for first-order methods. We consider the general problem of minimizing a nonsmooth Lipschitz convex function over a convex set while having access only to (very) noise estimates of its sub-gradients. We only assume that the noise possesses the first p moments, allowing for heavy-tailed regimes. In this setting, we provide the following contributions. First, we give the first optimal rates of convergence in expectation for the last iterate of clipped SsGM. These rates generalize to the bounded pth moment noise model all existing results for the standard finite-variance model. As a byproduct of our analysis of the last iterate, we provide improved rates for the average iterate. Second, assuming finite variance of the noise, we give the first optimal rates of convergence with high probability for a large class of averaging scheme of the iterates, including the standard average and a scheme that resembles the last iterate. Third, we show how to apply, via a kernel trick, the clipping operator to infinite dimensional input as those obtained by kernel methods, giving the first application of this approach to statistical learning. Finally, we provide empirical evidence supporting our theoretical
results
Factors associated with first-to-second-line attrition among patients with metastatic breast cancer in the real world
No full textBackground: Estimating patient attrition across lines of treatment (i.e. the probability that upon treatment failure the patient will not be able to receive a subsequent treatment) may be a valuable tool for optimizing treatment sequencing. We sought to describe the first-to-second-line attrition rate and factors associated with attrition in a real-world cohort of patients with metastatic breast cancer. Methods: The Gruppo Italiano Mammella (GIM)14/BIO-META (NCT02284581) is an ongoing, ambispective observational multicenter study enrolling patients with metastatic breast cancer receiving first-line therapy. In patients experiencing disease progression, attrition was defined as no further anticancer treatment and death within 6 months from the end of first-line therapy. The attrition rate from the first-to-second line was studied by descriptive analyses and univariate and multivariable logistic models were used to explore potentially predictive factors. Results: From January 2000 to December 2021, 3109 patients with metastatic breast cancer were enrolled in the GIM14/BIO-META study. Among them, 2498 patients experienced first-line treatment failure. Overall, first-to-second line attrition was 9.0% (95% confidence interval 7.9% to 10.1%), with similar attrition for patients with hormone receptor-positive/HER2-negative (8.5%) and HER2-positive (7.1%) breast cancer. Patients with triple-negative disease experienced the highest attrition (13.0%). Age, menopausal status, disease-free interval from primary tumor diagnosis, type of metastatic spread, and tumor subtype independently predicted first-to-second-line attrition. Conclusions: These findings could inform treatment decisions and guide clinical research on treatment sequencing. For instance, patients with the lowest risk of attrition may be the ideal candidates for trials exploring de-escalated first-line regimens, followed by more aggressive treatments upon progression
Endocrinology and Metabolic Diseases in Human Health
No full text: In 2022, we served as guest editors of the Nutrients Special Issue entitled "Endocrinology and Metabolic Diseases in Human Health". [...]
Giustizia Minorile. Ascesa e declino del tribunale per i minorenni in Italia
No full textIl Tribunale per i minorenni, istituito in epoca fascista e da allora giudice “specializzato” per tutte le cause civili, penali e amministrative che coinvolgono persone non ancora maggiorenni, sembra sull’orlo dell’inveramento del suo inevitabile destino. Nonostante l’ampia legittimazione ricevuta soprattutto negli anni Settanta del secolo scorso, questo tribunale è stato oggetto di ripetuti quanto fallimentari tentativi di riforma, culminati con il drastico ridimensionamento delle sue prerogative voluto dalla recente riforma Cartabia sul processo civile. Una débâcle che non ha avuto alcuna risonanza nel dibattito pubblico, dove i riflettori si sono nuovamente accesi sul problema dei giovani devianti e delle loro famiglie inadeguate. Questo libro ripercorre le vicende più significative dalle origini al declino
Increasing chemical coverage, accuracy, and reproducibility of the processing method for polar organic chemical integrative samplers
No full textInformation-Theoretic Detection of Bimanual Interactions for Dual-Arm Robot Plan Generation
No full textProgramming by demonstration is a strategy to simplify the robot programming process for non-experts via human demonstrations. However, its adoption for bimanual tasks is an underexplored problem due to the complexity of hand coordination, which also hinders data recording. This letter presents a novel one-shot method for processing a single RGB video of a bimanual task demonstration to generate an execution plan for a dual-arm robotic system. To detect hand coordination policies, we apply Shannon's information theory to analyze the information flow between scene elements and leverage scene graph properties. The generated plan is a modular behavior tree that assumes different structures based on the desired arms coordination. We validated the effectiveness of this framework through multiple subject video demonstrations, which we collected and made open-source, and exploiting data from an external, publicly available dataset. Comparisons with existing methods revealed significant improvements in generating a centralized execution plan for coordinating two-arm systems
Hybrid interacting quantum Hall thermal machine
No full textWe investigate a hybrid thermal machine based on a single closed quantum Hall edge channel forming a quantum dot. It is tunneling coupled with two quantum Hall states at nu = 2 in contact with reservoirs at different temperatures and chemical potentials. One of these edge states is also driven out of equilibrium by means of a periodic train of Lorentzian voltage pulses. This device allows to explore various possible working regimes including the engine, the heat pump, and the refrigerator configuration. Regions where two regimes coexist can also be identified. Moreover, the proposed setup exhibits robustness and in some parameter regions also slightly enhanced performance in the presence of electron-electron interactions
Rituximab retreatment guided by CD27+ B-cell count vs. clinical relapse in anti-MAG polyneuropathy: a cost-effective approach with lower cumulative doses
No full textIntroductionRituximab (RTX) is a widely used treatment for anti-MAG polyneuropathy, though standardized maintenance strategies are lacking. We aimed to compare two RTX retreatment protocols: (1) a full course (375 mg/m2/week for 4 weeks) administered at clinical relapse, and (2) a single infusion (375 mg/m2) at reappearance of peripheral CD27+ B cells—to evaluate their impact on disability progression over time.Patients and methodsWe retrospectively enrolled 29 patients with anti-MAG polyneuropathy, dividing them into two cohorts: (1) relapse (n = 19), treated with a full course at clinical relapse, or (2) Kim's protocol (n = 10), treated based on peripheral CD27+ B cell monitoring. Changes in INCAT, MRC sum score, and ISS from baseline to last follow-up were assessed.Results and discussionNo significant changes in MRC scores were observed in either cohort. Both cohorts showed a significant reduction in INCAT scores at last follow-up, with a tendency toward greater improvement in Kim's protocol cohort. ISS scores were significantly lower in Kim's protocol cohort compared to the relapse cohort (p < 0.01). Importantly, patients treated according to Kim's protocol received a cumulative RTX dose ~2.5 times lower than those treated upon relapse (p < 0.0001), despite showing comparable or better clinical outcomes.ConclusionA tailored maintenance strategy guided by peripheral CD27+ memory B-cell monitoring enables reduced cumulative RTX exposure while preserving clinical efficacy. This approach may improve cost-effectiveness and reduce treatment burden in patients with anti-MAG polyneuropathy