Making good predictions is a critical feature of decision making in situations such as investing and predicting the spread of diseases. Past literature indicates that people use recent and longer-term trends while making predictions. Nonetheless, less is known about how these factors affect how well people make predictions and the timing of their predictions. Further, identifying factors underlying predictive judgments could be an important behavioral factor in manic-depression, anxiety, substance use, age effects, and understanding how income inequality affects decision making. To understand how people make predictive judgments, we conducted two experiments. In Experiment 1, we used an investment task where participants had to predict the future price of a stock based on an exponential trend of information. We found that participants generally had lower earnings with steeper exponential trends (e.g. slower starting) and delayed their decisions to sell bad stocks with steeper trends. We extended these results in Experiment 2 with an updated task with exponential and inverse exponential trends. Overall, our results suggested that people delayed longer to make their prediction with slower starting exponential trends compared to faster starting inverse exponential trends and delayed their predictions longer with more linear trends compared to more trend trends. When deciding how long to explore, participants incorporated both the average trend and recent trend, though they shifted their responses depending on the overall functional form. These choices were ultimately biased to be optimistic or pessimistic based on whether the trend started fast or slow, respectively. Additionally, we found that participants who self-reported taking more gambling risk and depressive symptoms had a greater tendency to stay with faster starting trends and to leave with slower starting trends, suggesting they were even more optimistic given initially fast starting trends. Results pointing to an optimism bias based on the trend in information available to the participant could suggest that an aspect of sunk-cost fallacy is due to errors in predicting the likelihood of future success based on past information. Our findings help understand the dynamics of how people make predictive judgments over time and could inform future research into the mechanisms people use for prospective decision making. Additionally, future research and potential interventions could account for biases in how people perceive past trends to minimize harmful effects of sunk-cost fallacy when making predictions. / Psychology

Quantifying the properties of interest is an important problem in many domains, e.g., assessing the condition of a patient, estimating the risk of an investment or relevance of the search result. However, the properties of interest are often latent and hard to assess directly, making it difficult to obtain classification or regression labels, which are needed to learn a predictive models from observable features. In such cases, it is typically much easier to obtain relative comparison of two instances, i.e. to assess which one is more intense (with respect to the property of interest). One framework able to learn from such kind of supervised information is ranking SVM, and it will make a basis of our approach. Applications in bio-medical datasets typically have specific additional challenges. First, and the major one, is the limited amount of data examples, due to an expensive measuring technology, and/or infrequency of conditions of interest. Such limited number of examples makes both identification of patterns/models and their validation less useful and reliable. Repeated samples from the same subject are collected on multiple occasions over time, which breaks IID sample assumption and introduces dependency structure that needs to be taken into account more appropriately. Also, feature vectors are highdimensional, and typically of much higher cardinality than the number of samples, making models less useful and their learning less efficient. Hypothesis of this dissertation is that use of the functional norm regularization can help alleviating mentioned challenges, by improving generalization abilities and/or learning efficiency of predictive models, in this case specifically of the approaches based on the ranking SVM framework. The temporal nature of data was addressed with loss that fosters temporal smoothness of functional mapping, thus accounting for assumption that temporally proximate samples are more correlated. Large number of feature variables was handled using the sparsity inducing L1 norm, such that most of the features have zero effect in learned functional mapping. Proposed sparse (temporal) ranking objective is convex but non-differentiable, therefore smooth dual form is derived, taking the form of quadratic function with box constraints, which allows efficient optimization. For the case where there are multiple similar tasks, joint learning approach based on matrix norm regularization, using trace norm L* and sparse row L21 norm was also proposed. Alternate minimization with proximal optimization algorithm was developed to solve the mentioned multi-task objective. Generalization potentials of the proposed high-dimensional and multi-task ranking formulations were assessed in series of evaluations on synthetically generated and real datasets. The high-dimensional approach was applied to disease severity score learning from gene expression data in human influenza cases, and compared against several alternative approaches. Application resulted in scoring function with improved predictive performance, as measured by fraction of correctly ordered testing pairs, and a set of selected features of high robustness, according to three similarity measures. The multi-task approach was applied to three human viral infection problems, and for learning the exam scores in Math and English. Proposed formulation with mixed matrix norm was overall more accurate than formulations with single norm regularization. / Computer and Information Science

Les grands réseaux de communication sont partout, des centres de données avec des millions de serveurs jusqu’aux réseaux sociaux avec plusieurs milliards d’utilisateurs.Cette thèse est dédiée à l’étude fine de la complexité de différents problèmes combinatoires sur ces réseaux. Dans la première partie, nous nous intéressons aux propriétés des plongements des réseaux de communication dans les arbres. Ces propriétés aident à mieux comprendre divers aspects du trafic dans les réseaux (tels que la congestion). Plus précisément, nous étudions la complexité du calcul de l’hyperbolicité au sens de Gromov et de paramètres des décompositions arborescentes dans les graphes. Ces paramètres incluent la longueur arborescente (treelength) et l’épaisseur arborescente (treebreadth). Au passage, nous démontrons de nouvelles bornes sur ces paramètres dans de nombreuses classes de graphes, certaines d’entre elles ayant été utilisées dans la conception de réseaux d’interconnexion des centres de données. Le résultat principal dans cette partie est une relation entre longueur et largeur arborescentes (treewidth), qui est un autre paramètre très étudié des graphes. De ce résultat, nous obtenons une vision unifiée de la ressemblance des graphes avec un arbre, ainsi que différentes applications algorithmiques. Nous utilisons dans cette partie divers outils de la théorie des graphes et des techniques récentes de la théorie de la complexité / Large scale communication networks are everywhere, ranging from data centers withmillions of servers to social networks with billions of users. This thesis is devoted tothe fine-grained complexity analysis of combinatorial problems on these networks.In the first part, we focus on the embeddability of communication networks totree topologies. This property has been shown to be crucial in the understandingof some aspects of network traffic (such as congestion). More precisely, we studythe computational complexity of Gromov hyperbolicity and of tree decompositionparameters in graphs – including treelength and treebreadth. On the way, we givenew bounds on these parameters in several graph classes of interest, some of thembeing used in the design of data center interconnection networks. The main resultin this part is a relationship between treelength and treewidth: another well-studiedgraph parameter, that gives a unifying view of treelikeness in graphs and has algorithmicapplications. This part borrows from graph theory and recent techniques incomplexity theory. The second part of the thesis is on the modeling of two privacy concerns with social networking services. We aim at analysing information flows in these networks,represented as dynamical processes on graphs. First, a coloring game on graphs isstudied as a solution concept for the dynamic of online communities. We give afine-grained complexity analysis for computing Nash and strong Nash equilibria inthis game, thereby answering open questions from the literature. On the way, wepropose new directions in algorithmic game theory and parallel complexity, usingcoloring games as a case example

<p dir="ltr">Past research has extensively explored the benefits of acute aerobic exercise (AE) on memory and executive functions. Additionally, the cross-sectional relationship between muscle strength – a direct outcome of RE – and cognition is unknown, despite the simultaneous onset of muscle and cognitive decline in one’s thirties. However, the effects of acute resistance exercise (RE) on cognition remain understudied, despite the growing popularity of RE and evidence that RE may have distinct effects on cognition.. Therefore, the present study aimed to broaden our understanding of the connection between muscle strength and hippocampal-dependent memory and to investigate the influence of RE on memory and executive function.</p><p dir="ltr">A sample of 125 healthy young adults (18-50 years old) completed this study. On the first day of testing, subjects completed a cognitive battery testing aspects of hippocampal dependent memory, spatial abilities, and working memory, a maximal muscle strength testing session including handgrip strength and one-rep-max testing, and maximal aerobic capacity testing. Subjects completed a bioelectrical impedance assessment (BIA) body scan to measure body composition on Day 2. Day 3 consisted of a randomized controlled trial (RCT), where subjects completed either 42 minute moderate intensity RE (n = 62) or a seated rest (n = 61). Cognitive testing including a memory recognition task, an inhibitory control task, and a working memory task were performed both before and after the intervention. Subjects also completed lactate, blood pressure, and blood draw (only a subset of subjects (n = 59)) before and after intervention.</p><p dir="ltr">The results first revealed that after controlling for known covariates, those with greater handgrip strength performed better on mental rotation tasks (t = 2.14, p = 0.04, Δr2= 0.04), while those with higher upper-body relative strength did better on recognition (t = 2.78, p = 0.01, Δr2 = 0.06) and pattern separation (t = 2.03, p = 0.04, Δr2= 0.04) tasks. Further, while there was no acute effect of RE on memory performance, response times during measures of inhibitory control (t = 4.15, p < 0.01, d = 0.40) and working memory decreased after exercise (t = 7.01, p < 0.01, d = 0.46), along with decreases in P3 latency during the inhibitory control task (t =-5.99, p < 0.01, d = 0.58). Additionally, blood lactate (t =-17.18, p < 0.01, d = 2.06), serum brain derived neurotropic factor (BDNF) (t = -4.17, p < 0.01, d = 0.66), and systolic blood pressure (t = -10.58, p < 0.01, d = 0.99) all increased following RE, while diastolic blood pressure (t = 4.90, p < 0.01,d = 0.50) decreased. Notably, the change in systolic blood pressure (t = -2.83, p = 0.01, Δr2 = 0.06) was associated with improvements in behavioral measures of inhibitory control, changes in lactate (t = -2.26, p = 0.03, Δr2 = 0.04) and systolic blood pressure (t = -3.30, p < 0.01, Δr2 = 0.08) were also related to improved behavioral changes in working memory, and changes in lactate (t = -3.31, p < 0.01, Δr2= 0.08) and BDNF (t = -2.12, p = 0.04, Δr2= 0.08) related to faster P3 latency during inhibitory control. Importantly, these associations between physiological and cognitive changes were consistent across both exercise and rest groups, suggesting that physiological changes were linked to improved cognitive performance regardless of group assignment.</p><p dir="ltr">In conclusion, this study highlights the positive relationships between cross-sectional muscle strength and aspects of memory and spatial abilities, with distinct contributions from handgrip and upper body strength. Furthermore, acute RE was shown to enhance executive functions, particularly in terms of processing speed during inhibitory control (response time and P3 latency) and working memory (response time). This study suggests that RE can be a valid way to garner exercise-induced benefits on executive functions potentially through its influence on lactate, BDNF, and blood pressure, however, since these effects were evident regardless of intervention, more work is needed to determine if RE-induced changes have the same mechanisms. Overall, these findings underscore the potential benefits of muscle strength and RE on enhancing executive function in young and middle-aged adults.</p>