Browsing by Subject "Learning"

Now showing 1 - 20 of 32

Annual Report 2005
(MGT of America, 2005)
Annual Report 2006
(MGT of America, 2006)
Annual Report 2007
(MGT of America, 2007)
Change and reliability in the evolution of learning and memory.
(2009-05) Dunlap-Lehtilä, Aimee Sue
Why do animals learn to perform some behaviors while others are innate? Why do animals learn some things more easily than others? And, why do animals remember some things better than others? Theoreticians argue that patterns of environmental change explain these patterns, but we have little data to support these claims. I used statistical decision theory to model behaviors and fitness consequences, and experimental evolution studies with fruit flies where I manipulated patterns of environmental change across evolutionary time, to address the first two of these fundamental questions about the evolution of learning. The first experiment tested the effects of the reliability of experience and the fixity of the best action upon the evolution of learning and non-learning across 30 generations. I found that indeed, the interaction of these two variables determined when learning, and when non-learning evolved. The second study was a full factorial experiment manipulating the reliabilities of two modes of stimuli: olfactory and visual. After 40 generations, I found that as predicted, flies in environments where olfactory stimuli are reliable learned better about olfactory than color stimuli, with the same being true for color stimuli. Finally, I addressed the question of why animals remember some things better than others using a dynamic programming technique and experiment with blue jays, finding interactions between rates of change and time. These novel studies show the importance of reliability and change in evolution of learning and memory.
Context-dependent adaptation in the visual system
(2017-12) Mesik, Juraj
The visual system continuously adjusts its sensitivities to various visual features so as to optimize neural processing, a phenomenon known as adaptation. Although this rapid form of plasticity has been extensively studied across numerous sensory modalities, it remains unclear if its dynamics can change with experience. Specifically, the world we live in is composed of many different environments, or contexts, each of which contains its own statistical regularities. For example, forests contain more vertical energy and greenish hues than a desert landscape. Here we investigated the possibility that through experience, the visual system can learn statistical regularities in the visual input, and use this knowledge to adapt more quickly. In two sets of experiments, participants repeatedly adapted to previously unexperienced regularities in orientation statistics over the course of 3-4 sessions. They adapted either to rapidly presented sequences of oriented gratings containing orientation biases, or to natural visual input that was filtered to alter its orientation statistics. We found that experience did increase adaptation rate, but only in the experiments where participants adapted to a single set of altered statistics of natural input. We found no changes in adaptation rate in experiments where participants periodically switched between adapting to different statistical regularities. These results demonstrate that adaptation and experience can interact under some circumstances.
Contextual Influences on Cognitive and Psychophysiological Mechanisms of Learning in Early Adolescence
(2023-06) DeJoseph, Meriah
Learning is a central mechanism through which early experiences shape biological and behavioral development across the lifespan. One type of learning, called reinforcement learning, is posited to support youth’s ability to engage and adapt to their unique worlds with links to long-term social and emotional outcomes. Yet, individual differences in reinforcement learning across diverse environmental and experimental contexts remains poorly characterized in developmental samples. The current dissertation study integrated reinforcement learning and dynamic systems frameworks and drew upon newly adapted methodologies to capture how cognitive and psychophysiological processes of learning are modulated by socioemotional context. In a sample of 56 youth aged 12-15-years-old, this study leveraged a within-person experimental design and quantified continuous behavior and heart rate (~700 observations per system, per person) during an adapted reinforcement learning task with stimuli that varied in socioemotional relevance. Findings revealed that compared to traditionally-used benign or non-emotional stimuli, learning from stimuli high in socioemotional arousal enhanced behavioral performance. The use of computational modeling afforded valuable insights into the differential cognitive processes and strategies youth recruited to achieve such a behavioral advantage, demonstrating that socioemotional salience may have elicited faster value-updating processes and qualitative shifts in more exploitative decision-making. Underlying psychophysiological engagement seemed to be particularly modulated not by socioemotional salience as hypothesized, but by heightened sensitivity to learning from rewards, such that faster value-updating in the context of rewards aligned with more optimal psychophysiological flexibility and organization. Taken together, this study provides an important step in clarifying the contexts and modulatory processes that serve to enhance and support the unique ways youth learn and make decisions. Open questions remain about the adaptive utility of these various patterns of behavior, cognition, and psychophysiology across a variety of learning contexts, how they are shaped by prior lived experiences across development, and how they predict later psychosocial adjustment outcomes. Such work will shed light on how youth learn from–and adapt to–different contextual demands, with the potential to inform programs and policies that support youth’s ability to adjust to their dynamically changing ecologies.
Course specific collaborative teams in high school: an analysis of collaborative work, relationships and products.
(2011-03) Edwards, Daniel Lee
Improved student achievement has arguably always been a goal of schools and school districts. Within the past thirty years as the focus on increased student achievement has intensified, various calls for school reforms have resulted. These reform initiatives have taken on many appearances including government mandates as well as self-imposed changes. One of the most recent examples of reform that schools and school districts have embarked upon to bring about change has been the development of learning communities. Learning communities, often times referred to as Professional Learning Communities (PLC's), have evolved quite significantly over the past ten to fifteen years, often being implemented in a variety of different ways across all levels of education. An approach that has been often implemented at the high school level is the development of course-specific teams of teachers working together collaboratively on a variety of tasks associated with teaching. As approaches to the creation of learning communities have varied across settings, there is much to be learned by studying the application of these different approaches to the creation of a learning community and specifically teacher collaborative work that is focused at the course level. This research examined course-specific teams of teachers brought together for the purpose of working collaboratively to develop curriculum, instructional strategies, and assessments. Through the process of observing five course specific collaborative teams during two of their team meetings, follow up individual interviews with each member of the team, and an analysis of documents created by the team, this research illustrates the work of these teams. Three major findings inform the field of education related to the practice of learning communities and specifically teacher collaboration in a high school setting. First, collaboration that involves teachers with interdependent teaching roles, i.e. common courses, can result in improved professional practice. Second, having the opportunity to work collaboratively with teaching colleagues resulted in decreased feelings of isolation. Third, teacher collaboration resulted in improved relational trust among members of the collaborative teams.
Data Mining of Traffic Video Sequences
(University of Minnesota Center for Transportation Studies, 2009-09) Joshi, Ajay J.; Papanikolopoulos, Nikolaos
Automatically analyzing video data is extremely important for applications such as monitoring and data collection in transportation scenarios. Machine learning techniques are often employed in order to achieve these goals of mining traffic video to find interesting events. Typically, learning-based methods require significant amount of training data provided via human annotation. For instance, in order to provide training, a user can give the system images of a certain vehicle along with its respective annotation. The system then learns how to identify vehicles in the future - however, such systems usually need large amounts of training data and thereby cumbersome human effort. In this research, we propose a method for active learning in which the system interactively queries the human for annotation on the most informative instances. In this way, learning can be accomplished with lesser user effort without compromising performance. Our system is also efficient computationally, thus being feasible in real data mining tasks for traffic video sequences.
Does cerebellar cortex function as a forward internal model for motor control?
(2013-06) Hewitt, Angela L.
Motor control theorists have postulated that to produce rapid, finely tuned movements, a component of the control circuitry must bypass long sensory feedback delays by providing an estimate of the consequences resulting from a motor command. This control element, termed a forward internal model, receives an efferent copy of the motor command and information about the current state in order to predict the future state (i.e. kinematic variables like position, velocity) of the limb. Previous psychophysical, imaging, and patient case studies suggest that the cerebellum is a possible location for implementation of an internal model. However, very few electrophysiological studies have investigated whether the firing discharge from cerebellar neurons is consistent with the output of a forward internal model. To specifically evaluate the simple spike firing from Purkinje cells in lobules IV-VI, we trained rhesus macaques to perform different hand movement tasks using a 2 joint robotic manipulandum. Two electrophysiology experiments tested several aspects of a forward internal model. First, we hypothesize that Purkinje cell simple spike firing predicts future hand kinematics, even when the task is highly unpredictable. Second, the encoding is invariant, so that the model output can generalize to other tasks. A third hypothesis is that the simple spike discharge will show evidence of learning when animals adapt to a predictable mechanical perturbation, as expected from a forward internal model. Experimental results found many theoretical components of a forward internal model present in the Purkinje cell simple spike discharge. Simple spikes encode both feedforward and feedback representations of movement kinematics, with position and velocity signals explaining the most firing variability. These representations supply the predictive kinematic signals used downstream and the feedback information potentially used locally to construct predictions, calculate errors, and update the model. Many Purkinje cells exhibit dual encoding for a single kinematic parameter, so that these separate feedforward and feedback mechanisms may take place within individual cells. For most cells, model coefficients generated from random tracking data accurately estimate simple spike firing in either circular tracking or center-out reach. Adaptation to a predictable perturbation initiates steady, progressive changes in the parameter sensitivity (βs) of both the feedforward and feedback signals. The timing sensitivity (τ) also demonstrates significant shifts, with time encoding in the simple spikes often changing sign during adaptation (e.g. feedback to feedforward). Population analyses suggest that large changes in parameter sensitivity first occur in the feedback signals, then transfer to the feedforward representations. This may reflect use of the simple spike feedback to update model predictions. These results conclude that kinematic encoding from the cerebellar cortex uses a forward internal model that can generalize between tasks, but is also highly plastic and adaptable.
Effects of high-fidelity human patient simulation on self-efficacy, motivation and learning of first semester associate degree nursing students.
(2009-06) Kuznar, Kathleen A.
One of the newest methodologies in nursing education is high-fidelity human patient simulation (HPS). Many nursing educators have embraced the method as it offers a strategy to facilitate cognitive, affective, and psychomotor outcomes. Despite their popularity, however, HPS systems are costly and, in an era of cost containment and tuition increases, research must be employed to determine its effectiveness and guide its utilization. The purpose of this study is to determine how associate degree nursing students' self-efficacy, motivation, and learning in the simulated environment compare to nursing educational experiences without simulation. The mixed-method, quasi-experimental design was chosen for the study with a sample of first-semester associate degree nursing students at 2 technical colleges, 54 in the experimental group and 30 in the comparison group. Results indicated measures of self-efficacy and motivation increased throughout the semester for both groups. The simulation group had a statistically significant increase in general self-efficacy but no significant increase in nursing-specific academic and clinical self-efficacy. In contrast, the comparison groups had a significant increase in nursing academic self-efficacy but not in clinical or more general self-efficacy. Motivation measures were relatively consistent between the groups with only the measure of extrinsic motivation declining for the experimental group. When comparing the two groups on differences between pretest and posttest measures of self-efficacy and motivation, there were no significant differences. The experimental group scored significantly higher on the posttest knowledge examination. Results of interviews (n = 16) revealed specific themes, some unique to the simulation group and some common to members of both groups. The simulation students reported the importance of comprehensive skill practice, risk-free practice, group participation, and debriefing and instructor feedback. They were often able to identify a specific learning experience in the simulation lab that had impact on their practice. Technical skill knowledge was highly important for both groups. Students in both groups related the importance of a variety of courses in the first semester curriculum as increasing their nursing knowledge, self-efficacy and motivation. Simulation was found to be an acceptable learning strategy for novice associate degree nursing students.
Enhancement of learning: Does sleep benefit motor skill memory consolidation?
(2010-12) Borich, Michael Robert
Purpose: It remains unclear how the brain best recovers from neurologic injury and how to optimally focus rehabilitation approaches to maximize this recovery. Recent research has indicated that sleep may augment this recovery. Sleep has been shown to benefit memory consolidation for certain motor skills, but it remains unclear if this relationship exists for explicit, continuous, goal-directed motor skills with rehabilitation applications. We aimed to determine the neurobehavioral relationship between finger-tracking skill development and sleep following skill training in young, healthy subjects. Methods: Forty subjects were recruited to receive motor skill training in the morning (n=20) or the evening (n=20). Measures of skill and cortical excitability were collected before and after training. Following training, each group had a post-training interval consisting of waking activity or an interval containing sleep. After this twelve-hour interval, skill performance and cortical excitability were reassessed. Subjects underwent another twelve-hour interval containing either waking activity or a sleep episode and came back for a second assessment, twenty-four hours after training. A subset of subjects (n=10) underwent the same procedures except the training period involved simple, repeated movement of the finger. Results: Skill performance improved after training and then continued to improve offline during the first post-training interval. Improvement was not enhanced by sleep during this interval. Cortical excitability was not substantially altered by training but was related to level of skill performance at follow-up assessment. Sleep quality was also found to be related to level of skill at follow-up assessments. The skilled training period did not lead to significantly improved performance compared to simple movement activity. Discussion: These data suggest that sleep is not required for offline memory enhancement for a continuous, visuospatial finger-tracking skill. These findings are in agreement with recent literature indicating the type of motor skill trained may determine the beneficial effect of sleep on post-training information processing. These results, combined with related studies in patient populations, provide a foundation to evaluate the relationship between sleep, changes in neural activity, and the time course of continuous visuospatial motor skill learning in individuals following neurologic insult.
Essays in macroeconomic labor markets.
(2012-08) Michaud, Amanda Marie
In this thesis I study labor market dynamics in a macroeconomic context. The first chapter infers a theory of employment using the differences in wage and employment outcomes of job changers. This theory is used to understand differences in levels of unemployment and predict the effect of policy prohibiting employment discrimination against the unemployed. The second chapter examines the evolution of employment volatility relative to output in the US over the past half century. I find the increase is driven by certain demographic subgroups that can be thought of as highly skilled. I use this variation to see if a theory of increased skill transferability can account for the overall macro increase in relative employment volatility. The final chapter, joint with Jacek Rothert, proposes a link between government housing policy and savings in China. We construct a model of learning by doing in exports and find that optimal government policy restricting residential construction raises employment and output in the tradeable sector. This produces both a current account surplus and can be rationalized as benevolent because of the growth externality in learning by doing.
Essays on firm choice and international trade.
(2009-05) Lande, Katherine Nicole
This dissertation is composed of three essays that analyze the firm choice of how to service foreign destinations and which destinations to serve. In the first essay I provide a detailed decomposition of export growth at both the firm and product level. I first look at the export characteristics of Russian firm and the relationship between export growth margins and destination country characteristics. I then decompose product level export growth, focusing primarily on newly exported products, products withdrawn from the export market, and continuously exported products exported to new destinations. I show that there is a tendency for richer, larger countries to experience less growth on each of these margins than poorer, smaller countries and then discuss a model that accounts for these facts. Additionally, I show that even though many products are withdrawn from one or more destinations, very little export value is lost. I propose models which are consistent with the findings in the data.In the second essay, I show evidence that the geographic expansion of firm exports occurs slowly over time and that a large share of growth is due to continuing exporters entering new destinations. I also show that aggregated trade data can underestimate this value and hide the differing composition of export destinations among exporter types. New exporters enter large countries and destinations with characteristics similar to their domestic market. Less similar, distant or less developed countries are entered by firms already exporting to other destinations. I formulate a dynamic general equilibrium model to test if these patterns are due to firms learning how to export (as other recent empirical findings have suggested), or exogenous factors such as productivity growth. In this model, heterogeneous firms experience learning in the form of market entry costs that depend on export history. When calibrated to Russian firm level data, I find that learning plays a significant role in explaining the observed entry patterns, which standard trade models cannot account for. Additionally, by taking learning into account and targeting particular export destinations, governments can better assess the impact of liberalizations. Finally, in the third essay co-authored with Miguel F. Ricaurte, we use industry level data to show the striking differences among sectors in the ratios of exports to FDI sales. We determine what is needed to endogenously generate this pattern of export and FDI sales when firms make the decision to service a foreign market through either exports or foreign affiliates. By calibrating a model of monopolistically competitive firms, we find that tradability of goods is not enough to capture the observed sectoral differences, as is commonly assumed. We explore variants of the model and show that sector-specific taxes on multinationals and home bias allow us to replicate these differences.
Geoai: Challenges and Opportunities
(2020-05) Xie, Yiqun
Spatial data have tremendous value and are necessary components in many important societal applications. In recent years, our world has been witnessing a revolution brought by spatial technologies (e.g., Google Maps, Waze, Uber, Lyft, Grubhub, Lime, autonomous driving). According to a McKinsey Global Institute report, location data will generate about $600 billion annual revenue by 2020 with applications in energy, health, retail, etc. The world's economy also heavily relies on location and time data from over 2 billion GPS receivers, and these data are essential to many applications such as banks, airlines, police, emergency services, and telecommunications. Meanwhile, new types of spatial data are emerging at unprecedented scales and varieties (e.g., 25GB/hour per connected vehicle, 47.7PB per year by NASA by 2022). While spatial data are critical, valuable and collected at massive scales, they pose great challenges to traditional artificial intelligence (AI) techniques when applied to important societal problems. This thesis addresses three of these challenges. First, spatial data (e.g., crime or disease distribution, air quality) are often directly linked to our lived environments. As a result, decisions made on such data tend to have direct impacts on the life of citizens, and thus require statistical robustness to avoid errors which can have high economic and social costs (e.g., false alarm of a crime hotspot). Second, spatial data exhibit interdependency and variability, violating the common i.i.d. (identically and independently distributed) assumption in traditional statistics. This introduces new challenges to traditional optimization problems where spatial interdependency between nearby locations is often neglected and understudied (e.g., spatial contiguity required in land allocation). Finally, data and domain knowledge gaps are common in geospatial problems. For example, while Earth observation imagery is available in the tens of petabytes, there is very limited training data for many important objects or events (e.g., tree data for preventing fires and power blackouts) and expert knowledge is often required to create such data. This thesis investigates novel GeoAI techniques to explicitly address these challenges posed by spatial data and problems in three types of AI tasks: learning (i.e., unsupervised clustering); planning (i.e., spatial constraints and optimization); and perception (i.e., geospatial object mapping). First, the thesis proposes a significant DBSCAN approach for statistically-robust clustering to control the rate of spurious patterns. This work introduces a modeling of statistical significance for DBSCAN as well as a dual-convergence algorithm to speed up the computation. Second, the thesis proposes a fragmentation-free spatial allocation algorithm to explicitly model interdependency constraints among decision variables during optimization. Specifically, it introduces an optimization formulation with new spatial decision variables to model spatial contiguity and regularity constraints. It also proposes a hierarchical fragmentation elimination algorithm as well as a multi-layer integral image to efficiently solve the problem in a heuristic manner. Third, the thesis proposes a domain-knowledge assisted learning framework (i.e., TIMBER) to map geospatial objects (i.e., trees) with limited training data. The TIMBER framework introduces a geometric optimization formulation and a fast solver to generate candidates of tree-like structures for the deep learning model, which greatly reduces the difficulty of learning as well as the huge demand on training data. It also proposes a core object reduction algorithm to improve the computational performance. Extensive experiments and case studies show that the proposed approaches greatly outperform existing work in solution quality, and the proposed acceleration techniques greatly reduce the computational cost.
Learning by Doing and the Youth-Driven Model
(1998) Carlson, Stephan
In the non-formal setting of 4-H Youth Development, it has long been the motto that youth learn best when they are actively involved in relevant, real-world situations. This "learning by doing" is often associated with the type of learning model encouraged by 4-H organizations.
Learning To Communicate for Coordinated Multi-Agent Navigation
(2019-06) Hildreth, Dalton James
This work presents a decentralized multi-agent navigation approach that allows agents to coordinate their motion through local communication. Our approach allows agents to develop their own emergent language of communication through an optimization process that simultaneously determines what agents say in response to their spatial observations and how agents interpret communication from others to update their motion. We apply our communication approach together with the TTC-Forces crowd simulation algorithm and show a significant decrease in congestion and bottle-necking of agents, especially in scenarios where agents benefit from close coordination. In addition to reaching their goals faster, agents using our approach show coordinated behaviors including greeting, flocking, following, and grouping.Furthermore, we observe that communication strategies optimized for one scenario often continue to provide time-efficient, coordinated motion between agents when applied to different scenarios.This suggests that the agents are learning to generalize strategies for coordination through their communication “language".
Learning to sense sparse signals: simultaneous sensing matrix and sparsifying dictionary optimization
(University of Minnesota. Institute for Mathematics and Its Applications, 2008-05) Duarte-Carvajalino, Julio M.; Sapiro, Guillermo
The net-generation interior design student: an exploratory study assessing learning and engagement within a computer simulation environment.
(2009-06) Peterson, Julie Ellen
The first purpose of this experimental study was to determine if there were effects on achievement between traditional pencil-and-paper instructional strategies and computer simulated instructional strategies used to teach interior design business ethics. The second purpose was to determine the level of engagement of interior design students using traditional pencil-and paper instructional strategies compared to computer simulated instructional strategies to learn business ethics. The data offered both quantitative and qualitative evidence of preferred instructional strategies and what characteristics contributed towards level of engagement. Net-generation learners, born between 1982 and 2000, have been exposed to technology their whole lives and have come to expect the integration of various forms of multi-media instructional strategies within the classroom. Many studies have been conducted that integrate and analyze computer simulation and/or gaming with higher education, but research is very limited within the field of interior design. The study included 21 undergraduate interior design students. Analysis was both quantitative and qualitative in nature including descriptive statistics, frequencies, independent sample t-tests, ANCOVA statistical analysis, and questionnaires with both Likert-type and open-ended question formats. Even though statistical results were not found to be significant and were inconclusive, overall results indicated that the computer simulated case studies created an authentic, dynamic, and empowering learning environment that engaged the learners.
Neuroeconomic studies on personality and decision making
(2013-07) Hawes, Daniel R.
Neural activity causally underlies human cognition and behavior. Investigating the neurobiological principles and computational mechanisms governing brain activity during decision-making provides a way to improve theories of human behavior in the natural as well as social sciences (Glimcher & Rustichini 2004; Rustichini, 2009; Fehr & Rangel, 2009). In this context, the discipline of Neuroeconomics was originally conceived as an endeavor to interrogate neural activity during economic decision-making with the aim of evaluating competing decision theories (Rustichini, 2008; Glimcher, Camerer, Fehr & Poldrack 2009). From this origin, Neuroeconomics has evolved into a full-fledged enterprise of consilience; an attempt to not only test and bridge, but truly unify natural science and social science explanations of human behavior (Wilson, 1998; Glimcher & Rustichini, 2004; Rangel, Camerer & Montague, 2008).This dissertation binds two neuroeconomic studies of decision-making with an introduction and concluding commentary. The introduction presents a brief introduction to Neuroeconomics, meant to locate both research studies in the existing literature and philosophy of this field. The conclusion provides a brief appraisal of the role of Neuroeconomics in further advancing the kind of research into decision-making reported here. Both studies in this dissertation comprise investigations of human behavior during experience-based decision-making, with a special focus on the fundamental value computations that underlie such choice behavior.Study 1 investigates the role of neural reinforcement signals during learning of a strategic decision task from experience.Study 2 investigates the moderating effect of intelligence on neural reinforcement signals during a sequential binary choice task.Study 1 is reproduced from (Hawes, Vostroknutov & Rustichini 2013), and study 2 is reproduced from (Hawes, DeYoung, Gray & Rustichini; under review).
Participant Publications and Conference Presentations
(Bush Foundation and University of Minnesota, 2006) Center for Teaching and Learning

University Digital Conservancy

Browse by Subject

Browsing by Subject "Learning"