On the Generalization Ability of On-line Learning Algorithms

Cesa-Bianchi, Nicolò; Conconi, Alex; Gentile, Claudio

doi:10.7551/mitpress/1120.003.0051

Cited by 25 publications

(13 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The maximum error was chosen as the performance metric to emphasize reducing the error of outlying predictions, by grading the predictive ability of the model on its worst prediction. The choice of Bayesian optimization was based on its favorable comparison with similar methods, such as the grid search method (24).…”

Section: Optimization Of Hyperparameterssupporting

confidence: 93%

Fuzzy Inference Model for Railway Track Buckling Prediction

Słodczyk,

Fletcher,

Gitman

et al. 2023

Transportation Research Record: Journal of the Transportation R

View full text Add to dashboard Cite

The application of rail buckling models is often limited by uncertain information with respect to track properties, and many conventional models are poorly suited to network-wide or even regional application. Here, a methodology using fuzzy sets is presented that, when trained using buckling data can use inputs of track properties to predict the minimum buckling temperature increase for a particular track. An investigation of the impact of the size of training data and the influence of key track parameters on the minimum buckling temperature increase was conducted, and it was found that a high level of influence stems from the sleeper spacing and fastener torsional resistance parameters. The model was shown to give a low prediction error even for small dataset sizes of training data. The results of this work show the efficacy of a fuzzy sets based model when applied to track buckling prediction data, giving both a low error and rapid calculation times. The approach has potential for application for a wider array of variables, such as track geometry and vehicle dynamics, and is not limited to the study of track buckling owing to the flexibility of the fuzzy inference methodology.

show abstract

Section: Optimization Of Hyperparameterssupporting

confidence: 93%

Fuzzy Inference Model for Railway Track Buckling Prediction

Słodczyk,

Fletcher,

Gitman

et al. 2023

Transportation Research Record: Journal of the Transportation R

View full text Add to dashboard Cite

show abstract

“…Previous latent state models of behaviour have concentrated on animal behaviour in relatively well-studied task environments, such as courtship behaviours in fruit flies [17], visual detection in mice [16, 26, 18], or swim bouts in larval zebrafish [62]. These paradigms lend themselves well to models of latent states as the resultant observations are intuitively discretisable.…”

Section: Discussionmentioning

confidence: 99%

Unsupervised identification of internal perceptual states influencing psychomotor performance

Vardal,

Karapanagiotidis,

Stafford

et al. 2023

Preprint

View full text Add to dashboard Cite

When humans perform repetitive tasks over long periods, their performance is not constant. People may drift in and out of states that might be loosely categorised as engagement, disengagement or ‘flow’ and these will be reflected in multiple aspects of their performance (for example, reaction time, accuracy, criteria shifts and potentially longer-term strategy) but until recently it has been challenging to relate these behavioural states to the underlying neural mechanisms that generate them. Here, we took Magnetoencephalograpy recordings of participants performing an engaging task that required rapid, strategic behavioural responses. In this way we acquired both high density neural data and contemporaneous, dense behavioural data. Specifically, participants played a laboratory version of Tetris which collects detailed recordings of player input and game-state throughout performance. We asked whether it was possible to infer the presence of distinct behavioural states from the behavioural data and, if so, whether these states would have distinct neural correlates. We used hidden Markov modelling to segment the behavioural time series into states with unique behavioural signatures, finding that we could identify three distinct and robust behavioural states. We then computed occipital alpha power across each state. These within-participant differences in alpha power were statistically significant, suggesting that individuals shift between behaviourally and neurally distinct states during complex performance, and that visuo-spatial attention change across these states.

show abstract

“…The probabilistic choice is sometimes known as a "softmax" approach in reinforcement learning studies (Cesa-Bianchi et al, 2017) and the parameter α is a strength-of-response parameter. If α= 0 response is random, and as α grows the probability of choosing the action a with the highest r t a ð Þ increases.…”

Section: The Neural Autopilot Modelmentioning

confidence: 99%

A neural autopilot theory of habit: Evidence from consumer purchases and social media use

Camerer,

Xin,

Zhao

2023

J Exper Analysis Behavior

View full text Add to dashboard Cite

This article applies a two‐process “neural autopilot” model to field data. The autopilot model hypothesizes that habitual choice occurs when the reward from a behavior has low numerical “doubt” (i.e., reward prediction errors are small). The model toggles between repeating a previous choice (habit) when doubt is low and making a goal‐directed choice when doubt is high. The model has ingredients established in animal learning and cognitive neuroscience and is simple enough to make nonobvious predictions. In two empirical applications, we fit the model to field data on purchases of canned tuna and posting on the Chinese social media site Weibo. This style of modeling is called “structural” because there is a theoretical model of how different variables influence choices by agents (the “structure”), which tightly restricts how hidden variables lead to observed choices. There is empirical support for the model, more strongly for tuna purchases than for Weibo posting, relative to a baseline “reduced‐form” model in which current choices are correlated with past choices without a mechanistic (structural) explanation. An interesting set of predictions can also be derived about how consumers react to different kinds of changes in prices and qualities of goods (this is called “counterfactual analysis”).

show abstract

On the Generalization Ability of On-line Learning Algorithms

Cited by 25 publications

References 44 publications

Fuzzy Inference Model for Railway Track Buckling Prediction

Fuzzy Inference Model for Railway Track Buckling Prediction

Unsupervised identification of internal perceptual states influencing psychomotor performance

A neural autopilot theory of habit: Evidence from consumer purchases and social media use

Contact Info

Product

Resources

About