Low Learning Rate Research Articles

Evidence for age-associated cognitive decline from Internet game scores

IntroductionLumosity's Memory Match (LMM) is an online game requiring visual working memory. Change in LMM scores may be associated with individual differences in age-related changes in working memory. MethodsEffects of age and time on LMM learning and forgetting rates were estimated using data from 1890 game sessions for users aged 40 to 79 years. ResultsThere were significant effects of age on baseline LMM scores (β = −.31, standard error or SE = .02, P < .0001) and lower learning rates (β = −.0066, SE = .0008, P < .0001). A sample size of 202 subjects/arm was estimated for a 1-year study for subjects in the lower quartile of game performance. DiscussionOnline memory games have the potential to identify age-related decline in cognition and to identify subjects at risk for cognitive decline with smaller sample sizes and lower cost than traditional recruitment methods.

When will wind energy achieve grid parity in China? – Connecting technological learning and climate finance

China has adopted an ambitious plan for wind energy deployment. This paper uses the theory of the learning curve to investigate financing options to support grid parity for wind electricity. First, relying on a panel dataset consisting of information from 1207 wind projects in China’s thirty provinces over the period of 2004–2011, this study empirically estimates the learning rate of onshore wind technology to be around 4.4%. Given this low learning rate, achieving grid parity requires a policy of pricing carbon at 13€/ton CO2e in order to increase the cost of coal-generated electricity. Alternatively, a learning rate of 8.9% would be necessary in the absence of a carbon price. Second, this study assesses the evolution of additional capital subsidies in a dynamic framework of technological learning. The implicit average CO2 abatement cost derived from this learning investment is estimated to be around 16€/ton CO2e over the breakeven time period. The findings suggest that climate finance could be structured in a way to provide up-front financing to support this paradigm shift in energy transition.

Object detection in video sequences by a temporal modular self-adaptive SOM

A video segmentation algorithm that takes advantage of using a background subtraction (BS) model with low learning rate (LLR) or a BS model with high learning rate (HLR) depending on the video scene dynamics is presented in this paper. These BS models are based on a neural network architecture, the self-organized map (SOM), and the algorithm is termed temporal modular self-adaptive SOM, TMSA_SOM. Depending on the type of scenario, the TMSA_SOM automatically classifies and processes each video into one of four different specialized modules based on an initial sequence analysis. This approach is convenient because unlike state-of-the-art (SoA) models, our proposed model solves different situations that may occur in the video scene (severe dynamic background, initial frames with dynamic objects, static background, stationary objects, etc.) with a specialized module. Furthermore, TMSA_SOM automatically identifies whether the scene has drastically changed (e.g., stationary objects of interest become dynamic or drastic illumination changes have occurred) and automatically detects when the scene has become stable again and uses this information to update the background model in a fast way. The proposed model was validated with three different video databases: Change Detection, BMC, and Wallflower. Findings showed a very competitive performance considering metrics commonly used in the literature to compare SoA models. TMSA_SOM also achieved the best results on two perceptual metrics, Ssim and D-Score, and obtained the best performance on the global quality measure, FSD (based on F-Measure, Ssim, and D-Score), demonstrating its robustness with different and complicated non-controlled scenarios. TMSA_SOM was also compared against SoA neural network approaches obtaining the best average performance on Re, Pr, and F-Measure.

Technical, Perceptual and Motor Skills in Novice-Expert Water Polo Players

The 4 tasks (A, B, C, and Y) have the characteristic of containing one more element than the task performed before it. In fact, task B introduces the slalom which is not present in task A. Task C introduces the ball control that are not present in tasks A and B, whereas task Y introduces the slalom and ball control in a visual dual task situation developed in horizontal swimming over a distance of 20 m at maximum speed. This exercise not included in task C. These tasks were performed by a group of pre-adolescent players and national under 18 water polo players. The novice players showed that tasks B and C are predictors of task Y. Such characteristics were not present in the expert players. The novice players also had difficulty in performing task Y because of the visual-attention overload, a difficulty that was not present in the expert players. To improve the 4 skills, the coach of the novice players developed a technical-didactic program, which was checked 6 months after the pretest. The posttest was not significantly different from the pretest while the individual discriminant analysis identified the improvements in some novice players, which on elaboration proved significant, enabling us to distinguish 2 subgroups, one with higher learning rates and the other with lower learning rates. In the practical applications, we describe the didactic tools (task analysis) and the different levels of development of technical skills in water polo. Improvements in these skills are explained through computational models like the HMOSAIC (Hierarchical, Modular, Selection and Identification for Control) while the individual discriminant analysis enables us to do a longitudinal analysis that is not possible with cross-sectional models.

Application of Fault Diagnosis for Air Blower Based on Genetic Fuzzy Neural Network

The way of fault characteristic parameters fuzzy processing and optimizing the weights and thresholds of ANN by GA are studied. As a result, the convergent rate and convergent precision are greatly increased. Application to the fault diagnosis of a air blower system shows the new model overcomes the low learning rate and local optima of BP algorithm, and the fault diagnosis precision is effectively improved.

Genetic variation in dopamine‐related gene expression influences motor skill learning in mice

Several neurodevelopmental disorders with a strong genetic basis, including attention-deficit/hyperactivity disorder, autism spectrum disorders and developmental coordination disorder, involve deficits in fine motor skills. This phenotype may depend on heritable variation in components of the dopamine (DA) system, which is known to play a critical role in motor skill learning. In this study, we took advantage of two inbred strains of mice (BALB/c and C57BL/6) that differ markedly in the number of midbrain DA neurons in order to investigate the influence of such naturally occurring genetic variation on the acquisition and performance of fine motor skills. Gene expression analysis of midbrain, frontal cortex and striatum showed significant differences in the expression of presynaptic and postsynaptic dopaminergic (DAergic) markers (e.g. tyrosine hydroxylase, DA transporter, DA D4 receptor, DA D5 receptor and DARPP-32) between these two strains. BALB/c mice had lower learning rate and performance scores in a complex skilled reaching task when compared with C57BL/6 mice. A negative correlation was found between the motor learning rate and level of DARPP-32 mRNA expression in the frontal cortex contralateral to the trained forelimb. The rate of motor learning was also negatively correlated with the levels of DARPP-32 and DA D1 receptor mRNAs in the striatum. Our results suggest that genetically driven variation in frontostriatal DAergic neurotransmission is a major contributor to individual differences in motor skill learning. Moreover, these findings implicate the D1R/cAMP/DARPP-32 signaling pathway in those neurodevelopmental disorders that are associated with fine motor skill deficits.

Hybrid Optimized Back propagation Learning Algorithm for Multi-layer Perceptron

Standard neural network based on general back propagation learning using delta method or gradient descent method has some great faults like poor optimization of error-weight objective function, low learning rate, instability .This paper introduces a hybrid supervised back propagation learning algorithm which uses trust-region method of unconstrained optimization of the error objective function by using quasi-newton method .This optimization leads to more accurate weight update system for minimizing the learning error during learning phase of multi-layer perceptron.[13][14][15] In this paper augmented line search is used for finding points which satisfies Wolfe condition. In this paper, This hybrid back propagation algorithm has strong global convergence properties & is robust & efficient in practice.

Fault Diagnosis Analysis of Rotor System Based on RBF Neural Network and Dynamic Systems

The RBF network was applied in the rotor system to realize the fault diagnosis aiming the mapping complexity between fault symptoms and fault patterns. It can overcome the problems of low learning rates of convergence and falling easily into part minimums in BP algorithm, and improve the precision of diagnosis. The normalized values of seven frequency ranges in amplitude spectrum were used as the fault characteristic quantity, the RBF network was trained to diagnose the faults of rotor system. The results show that RBF neural network is a valid method of diagnosis of mechanical failure.

Learning rates and future cost curves for fossil fuel energy systems with CO2 capture: Methodology and case studies

The broadly applicable equations for estimating learning rates of cost variables in energy systems with CO2 capture (CC) are formulated, in which the effect of overall plant efficiency upgrade on learning rates is reflected. Based on the equations, as a case study, we estimate the learning rates, predict the future cost trend of IGCC power plants with CC in China, and examine the effect of plant efficiency upgrade on its future cost. It is revealed that the learning rates of the whole CC plant are relevant not only to the learning rate of each subunit, but also to its cost proportion and the overall plant efficiency upgrade. Results from case study show that the learning rates of IGCC+CC in China are in the range of 0.0964–0.2022 for unit investment, 0.0726–0.1489 for COE, and 0.0636–0.1462 for cost of CO2 avoidance (COA). When the cumulative production reaches 100GW, the investment for IGCC+CC will decrease from the current level (approximately 2150$/kW) to around 760–1350$/kW, COE will decrease to 46–68$/MWh, and COA will fall from 33.4$/t to 16–25$/t. Sensitivity analysis indicates that overall plant efficiency upgrade and the capacity at which learning begins pose significant effects on cost reduction. Compared with PC+CC, for gradual learning with a low learning rate, the unit investment of IGCC+CC will be a little bit higher than that of PC+CC in the future. For rapid and moderate learning, IGCC+CC will be more expensive than PC+CC in the near term, while breakeven points are observed with the cumulative experiences growing, indicating that IGCC+CC can economically perform better than PC+CC in the medium and long term. The paper provides an approach to estimate the learning rates of CC plants, and thus to project their future cost curves, which will help to formulate the first clear-cut CCS roadmap in China and to aid the identification of key CC technologies that should be focused on.

CEOs in Family Firms: Does Junior Know What He's Doing?

We model the evolution of CEO quality in family firms. Agents learn about their own qualities over time by observing the successful outcome of their own actions, while failures drive agents out of the market. We show that joining the family business, while reducing the probability of failure, impairs the learning process. This implies that the pool of family CEOs, shielded from failure and with lower learning rates, is endogenously worse than that of outside CEOs in the long-run. Our model also produces an even/odd fluctuation in intergenerational quality which is consistent with existing empirical literature. Finally, we show that an increase in bankruptcy costs and capital costs increase the probability an heir joins the family firm.

A reinforcement learning approach to instrumental contingency degradation in rats

Goal-directed action involves a representation of action consequences. Adapting to changes in action-outcome contingency requires the prefrontal region. Indeed, rats with lesions of the medial prefrontal cortex do not adapt their free operant response when food delivery becomes unrelated to lever-pressing. The present study explores the bases of this deficit through a combined behavioural and computational approach. We show that lesioned rats retain some behavioural flexibility and stop pressing if this action prevents food delivery. We attempt to model this phenomenon in a reinforcement learning framework. The model assumes that distinct action values are learned in an incremental manner in distinct states. The model represents states as n-uplets of events, emphasizing sequences rather than the continuous passage of time. Probabilities of lever-pressing and visits to the food magazine observed in the behavioural experiments are first analyzed as a function of these states, to identify sequences of events that influence action choice. Observed action probabilities appear to be essentially function of the last event that occurred, with reward delivery and waiting significantly facilitating magazine visits and lever-pressing respectively. Behavioural sequences of normal and lesioned rats are then fed into the model, action values are updated at each event transition according to the SARSA algorithm, and predicted action probabilities are derived through a softmax policy. The model captures the time course of learning, as well as the differential adaptation of normal and prefrontal lesioned rats to contingency degradation with the same parameters for both groups. The results suggest that simple temporal difference algorithms with low learning rates can largely account for instrumental learning and performance. Prefrontal lesioned rats appear to mainly differ from control rats in their low rates of visits to the magazine after a lever press, and their inability to initially detect weak contingency changes.

Wind power learning rates: A conceptual review and meta-analysis

In energy system models endogenous technological change can be introduced by implementing so-called technology learning rates specifying the quantitative relationship between the cumulative experience of a technology and its cost. The objectives of this paper are to: (a) provide a conceptual review of learning curve model specifications; and (b) conduct a meta-analysis of wind power learning rates. This permits an assessment of a number of important specification and data issues that influence these learning rates. The econometric analysis builds on 113 estimates of the learning-by-doing rate presented in 35 studies. The meta-analysis indicates that the choice of the geographical domain of learning, and thus the assumed presence of learning spillovers, is an important determinant of wind power learning rates. We also find that the use of extended learning curve concepts, e.g., integrating public R&D effects, appears to result in lower learning rates than those generated by so-called single-factor learning curve studies. Overall the empirical findings suggest that future studies should pay increased attention to the issue of learning and knowledge spillovers in the renewable energy field, as well as to the interaction between technology learning and R&D efforts.

Natural variation in learning and memory dynamics studied by artificial selection on learning rate in parasitic wasps

Animals form memory types that differ in duration and stability. The initial anaesthesia-sensitive memory (ASM) can be replaced by anaesthesia-resistant memory (ARM), and/or by protein synthesis-dependent, long-term memory (LTM). We previously showed that two closely related parasitic wasp species differ in learning rate and memory consolidation. In Cotesia glomerata, LTM lasting at least 24 h was formed after single-trial conditioning, whereas single-trial conditioning led to ARM that waned before 24 h in Cotesia rubecula. This species formed LTM only after repeated conditioning trials spaced in time. Here, we used artificial selection on learning rate to investigate whether selection for a low learning rate in C. glomerata would result in C. rubecula-like memory dynamics. Memory consolidation was tested by using cold-shock anaesthesia and protein synthesis inhibitors. After single-trial conditioning, ARM was consolidated within hours in unselected C. rubecula, but directly, without an intermediate ARM phase, into LTM in unselected C. glomerata. We obtained low learning rate selection lines of C. glomerata wasps that, like C. rubecula, did not form LTM after single-trial conditioning, to see whether such wasps would then consolidate ARM instead of LTM. We showed that this was not the case. The selected wasps formed LTM after repeated, spaced conditioning trials, but formed only ASM without consolidation of ARM or LTM after single-trial learning. Ecological consequences of this type of memory formation are discussed.

Fuzzy Classification Method for Processing Incomplete Dataset

Pattern classification is one of the most important topics for machine learning research fields. However incomplete data appear frequently in real world problems and also show low learning rate in classification models. There have been many researches for handling such incomplete data, but most of the researches are focusing on training stages. In this paper, we proposed two classification methods for incomplete data using triangular shaped fuzzy membership functions. In the proposed methods, missing data in incomplete feature vectors are inferred, learned and applied to the proposed classifier using triangular shaped fuzzy membership functions. In the experiment, we verified that the proposed methods show higher classification rate than a conventional method.

On the Weight Convergence of Elman Networks

An Elman network (EN) can be viewed as a feedforward (FF) neural network with an additional set of inputs from the context layer (feedback from the hidden layer). Therefore, instead of the offline backpropagation-through-time (BPTT) algorithm, a standard online (real-time) backpropagation (BP) algorithm, usually called Elman BP (EBP), can be applied for EN training for discrete-time sequence predictions. However, the standard BP training algorithm is not the most suitable for ENs. A low learning rate can improve the training of ENs but can also result in very slow convergence speeds and poor generalization performance, whereas a high learning rate can lead to unstable training in terms of weight divergence. Therefore, an optimal or suboptimal tradeoff between training speed and weight convergence with good generalization capability is desired for ENs. This paper develops a robust extended EBP (eEBP) training algorithm for ENs with a new adaptive dead zone scheme based on eEBP training concepts. The adaptive learning rate and adaptive dead zone optimize the training of ENs for each individual output and improve the generalization performance of the eEBP training. In particular, for the proposed eEBP training algorithm, convergence of the ENs' weights with the adaptive dead zone estimates is proven in the sense of Lyapunov functions. Computer simulations are carried out to demonstrate the improved performance of eEBP for discrete-time sequence predictions.

Similarity and Rules United: Similarity‐ and Rule‐Based Processing in a Single Neural Network

A central controversy in cognitive science concerns the roles of rules versus similarity. To gain some leverage on this problem, we propose that rule- versus similarity-based processes can be characterized as extremes in a multidimensional space that is composed of at least two dimensions: the number of features (Pothos, 2005) and the physical presence of features. The transition of similarity- to rule-based processing is conceptualized as a transition in this space. To illustrate this, we show how a neural network model uses input features (and in this sense produces similarity-based responses) when it has a low learning rate or in the early phases of training, but it switches to using self-generated, more abstract features (and in this sense produces rule-based responses) when it has a higher learning rate or is in the later phases of training. Relations with categorization and the psychology of learning are pointed out.

Testing for the presence of some features of increasing returns to adoption factors in energy system dynamics: An analysis via the learning curve approach

The purpose of this paper is to explain the sources of energy system lock-in. It presents a comparative analysis of the respective contributions of some features of increasing returns to adoption factors, i.e. learning-by-doing, learning-by-searching and returns to scale effects in explaining the technological change dynamics in the energy system. The paper is technically based on a critical analysis of the learning curve approach. Econometric estimation of learning and scale effects inherent to seven energy technologies were performed by the use of several learning curve specifications. These specifications permit to deal with some crucial issues related to the learning curve estimation which are associated with the problem of omitted variable bias, the endogeneity effects and the choice of learning indicators. Results show that dynamic economies from learning effects coupled with static economies from scale effects are responsible for the lock-in phenomena of the energy system. They also show that the magnitude of such effects is correlated with the technology life cycle (maturity). In particular, results point out that, 1) the emerging technologies exhibit low learning rates associated with diseconomies of scale which are argued to be symptomatic of the outset of the deployment of new technologies characterized by diffusion barriers and high level of uncertainty, 2) the evolving technologies present rather high learning rates meaning that they respond quickly to capacity expansion and R&D activities development, 3) conventional mature technologies display low learning rates but increasing returns to scale implying that they are characterized by a limited additional diffusion prospects.

Dynamic behavioural changes in the Spontaneously Hyperactive Rat: 1. Control by place, timing, and reinforcement rate

Several existing theoretical models of Attention Deficit Hyperactivity Disorder (ADHD) predict that a high or low learning rate contributes to some cases of ADHD; and that this is controlled by the dopamine signal. The Spontaneously Hypertensive (Hyperactive) Rat has an abnormal Dopamine Transporter (DAT), which is likely to be hypofunctional, thereby increasing the dopamine concentration and the learning rate. We therefore test the prediction that the SHR learns too fast. Using a variable interval reinforcement schedule, we looked for increased responding within each interval, at around the durations of the last three intervals. At the time predicted by the previous interval, SHR responded significantly (10%) more than their baseline rate, and control WKY rats similarly less than their baseline. At the time predicted by the second (and third) previous intervals, there were minor trends (and no change) in the same direction. In summary, at the times predicted, SHR respond more than control rats. The effect size is insufficient to account for all the differences from control rats, such as their nearly threefold greater operant responding. At the times when SHR increase responding, WKY suppress it. This demonstrates similar learning to the SHR but the opposite response, probably due to anxiety about the bang accompanying reinforcer delivery: WKY are known to have a very negative response to stress. Interval schedules with randomised interval lengths have considerable potential for assessing multiple causes of behaviour, particularly those acting over short timecourses. Alterations in learning rate are not a sufficient explanation for either SHR or ADHD difference from controls.

An impoverished machine: Challenges to human learning and instructional technology

Many of the limitations to human learning and processing identified by cognitive psychologists over the last 50 years still hold true, including computational constraints, low learning rates, and unreliable processing. Instructional technology can be used in classrooms and in other learning contexts to address these limitations to learning. However, creating technological innovations is not enough. As part of psychological science, the development and assessment of instructional systems should be guided by theories and practices within the discipline. The technology we develop should become an object of research like other phenomena that are studied. In the present article, I present an informal account of my own work in assessing instructional technology for engineering thermodynamics to show not only the benefits, but also the limitations, in studying the technology we create. I conclude by considering several ways of advancing the development of instructional technology within the SCiP community, including interdisciplinary research and envisioning learning contexts that differ radically from traditional learning focused on lectures and testing.

Performance of serial assembly line designs under unequal operator speeds and learning

The traditional balanced assembly line designs can perform quite inefficiently under the presence of high labor turnover, low operator learning rates, and stochastic processing times. In these situations assembly line designs based on dynamic work allocation and work sharing principles have been shown to render a higher and less variable throughput. However, for situations where low labor turnover conditions and high operator learning rates exist, the traditional balanced lines still tend to be most productive. This paper has two objectives: to introduce a method based on work sharing principles, which we denominate Modified Work Sharing (MWS), and to develop some simple analytical tools that will allow us to compare the performance of this method with the traditional and other dynamic work allocation line designs. These results suggest that the traditional method is the most affected by the introduction of new operators in the production lines and thus the most affected by variability in general. On the other hand, dynamic work allocation methods appear to better absorb the variability introduced by new operators by sharing the workload of new operators with the more experienced members of the line.