МЕТОД НАВЧАННЯ БЕЗ ВЧИТЕЛЯ ІЄРАРХІЧНОГО ЕКСТРАКТОРА ВІЗУАЛЬНИХ ОЗНАК НА ОСНОВІ МОДИФІКАЦІЇ НЕЙРОННОГО ГАЗУ
Bibliographic record
Abstract
The modern technologies of the intellectual analysis of visual information for solving the problem of unsupervised training in real time with the aim of adapting to unknown conditions of observation are analyzed. It is proposed to use 10 layers of the well-known neural network VGG-16 as a model of the hierarchical extractor of visual features that can be used in the transfer learning tasks. The use of the principles of the neural gas to increase the convergence rate of the algorithm of usupervised learning of the extractor of visual features under the conditions of a limited amount of training data is considered. The modification of the neuron gas aimed to sparse coding of input observations is based on the optimized orthogonal matching pursuit algorithm that was used to increase the informativeness of the feature set in condition of limited sample size. Training dataset is generated by selecting from a popular image base ImageNet and selecting patches from selected images or feature maps on a given layer. The method of so-called information-extreme machine learning of decision rules is proposed for assessing the efficiency of the proposed feature extractor. Information-extreme learning is based on the use of binary coding of the feature representation of observations and the construction of radial-basic decision rules in Hamming's binary space. The implementation of the algorithm is based on the use of computationally simple operations such comparation with threshold and a bitwise XOR. Optimization of the geometric parameters of the partition feature space into separated classes is carried out in the binary space, therefore, it can be implemented by the method of a sequential direct busting with a given step, since such steps are relatively small. For optimizing parameters of encoding observations rules is used population-based particle swarm algorithm for searching global maximum of logarithmic information Kullback’s criterion in admissible domain of it function. In this case we normalized modification information criterion which is function of the first and second kind errors is used. The effectiveness of training of decision rules in the case of the use of an extractor supervise trained with by a stochastic gradient descent method, with case of supervised trained feature extractor is compared. According to the results of physical modeling unsupervised learning of extractor ensures the accuracy of decisive rules to 96.4% which is inferior to the accuracy of supervised learning which is equal to 98.7% are shown.
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 imitationNot 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.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.035 | 0.014 |
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.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
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".