MétaCan
Menu
← Back to cohort
Record W7033362402

A rank-determination method for the spectral learning of stochastic weighted automata

2019· dissertation· en· W7033362402 on OpenAlexaff

Bibliographic record

VenueeScholarship@McGill (McGill) · 2019
Typedissertation
Languageen
FieldEnvironmental Science
TopicAquatic Invertebrate Ecology and Behavior
Canadian institutionsMcGill University
Fundersnot available
KeywordsRank (graph theory)Singular value decompositionNoise (video)Sequence (biology)Singular valueMatrix (chemical analysis)Matrix decompositionProbabilistic logic
DOInot available

Abstract

fetched live from OpenAlex

In this work, a rank-determination method for the spectral learning algorithms of stochastic weighted finite automata (WFA) is proposed.In the spectral learning of stochastic WFA, a hyper-parameter which is the rank of the model needs to be specified by a user.However, the user usually does not know the true rank in advance.Therefore, some cross-validation process is employed for choosing the rank.This method is inefficient since it requires to learn multiple models.Our method can estimate the rank of the model during a single spectral learning process, and hence it eliminates the need for the cross-validation process.The idea is that when the noise level in data is low, analyzing the noise level and comparing it with the singular values of the sub-Hankel matrix might give us some insight into the number of effective singular values.Because the rank of any complete sub-Hankel matrix is equal to the number of state in a stochastic WFA, we can approximately estimate the number of state in a stochastic WFA by estimating the rank of a complete sub-Hankel matrix.To achieve this, we replace the deterministic singular value decomposition (SVD) with the fixed-precision randomized singular value decomposition (RSVD) in the spectral learning algorithm and propose two methods to estimate the noise level in data.By combining the noise-estimation method and the modified spectral learning algorithm, we obtain the new method, noise estimation based rank revealing method (NERR).Experimental results on datasets of the probabilistic automata learning competition (Pautomac) and the sequence prediction challenge (SPiCe) demonstrate the utility of this method.v Résumé Dans ce travail, une méthode de détermination de rang pour les algorithmes d'apprentissage spectral des automates finis pondérés (WFA) stochastiques est proposée.Pour l'apprentissage spectral des automates finis pondérés stochastiques, un hyperparamètre, qui correspond au rang du modèle, doit être spécifié par l'utilisateur.Cependant, l'utilisateur ne connaît généralement pas le rang réel à l'avance.Par conséquent, un processus de validation croisée est utilisé pour choisir le rang.Cette méthode est inefficace car elle nécessite l'apprentissage de plusieurs modèles.Notre méthode permet d'estimer le rang du modèle au cours d'un processus d'apprentissage spectral unique, éliminant ainsi la nécessité du processus de validation croisée.L'idée est que, lorsque le niveau de bruit dans les données est faible, l'analyse du niveau de bruit et sa comparaison avec les valeurs singulières de la matrice sous-Hankel pourraient nous donner une idée du nombre de valeurs singulières effectives.Étant donné que le rang de toute matrice sous-Hankel complète est égal au nombre d'états dans un automate fini pondéré stochastique, nous pouvons estimer approximativement le nombre d'états dans un automate fini pondéré stochastique en estimant le rang d'une matrice sous-Hankel complète.Pour ce faire, nous remplaçons la décomposition en valeurs singulières (SVD) d'ordre déterministe par la décomposition en valeurs singulières d'ordre aléatoire (RSVD) à précision fixe dans l'algorithme d'apprentissage spectral et proposons deux méthodes d'estimation du niveau de bruit dans les données.En combinant la méthode d'estimation du bruit et l'algorithme d'apprentissage spectral modifié, nous obtenons la nouvelle méthode, la méthode de révélation de rang basée sur l'estimation du bruit (NERR).Les résultats expérimentaux sur les jeux de données du concours d'apprentissage des automates probabilistes (Pautomac) et du défi de prédiction de séquence (SPiCe) démontrent l'utilité de cette méthode.vii Acknowledgements Foremost, I would like to express my sincere gratitude to my thesis advisors Prof. Xiao-Wen Chang and Prof. Doina Precup for the financial support of my master study and research, for their patience, motivation, enthusiasm, and immense knowledge.Their guidance helped me in all the time of research and writing of this thesis.I could not have imagined having better

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.004
metaresearch head score (Gemma)0.019
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Theoretical or conceptual · Consensus signal: none
GenreCandidate signal: Methods · Consensus signal: Methods
Teacher disagreement score0.004
Threshold uncertainty score0.023

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0040.019
Meta-epidemiology (narrow)0.0020.001
Meta-epidemiology (broad)0.0020.002
Bibliometrics0.0020.001
Science and technology studies0.0010.002
Scholarly communication0.0020.004
Open science0.0020.003
Research integrity0.0020.004
Insufficient payload (model declined to judge)0.0040.002

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.014
GPT teacher head0.268
Teacher spread0.253 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designTheoretical or conceptual
Domainnot available
GenreMethods

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2019
Admission routes1
Has abstractyes

Explore more

Same venueeScholarship@McGill (McGill)→Same topicAquatic Invertebrate Ecology and Behavior→French-language works237,207→