An official website of the United States government
Bayesian Knowledge Tracing (BKT) is a commonly used approach for student modeling, and Long Short Term Memory (LSTM) is a versatile model that can be applied to a wide range of tasks, such as language translation. In this work, we directly compared three models: BKT, its variant Intervention-BKT (IBKT), and LSTM, on two types of student modeling tasks: post-test scores prediction and learning gains prediction. Additionally, while previous work on student learning has often used skill/knowledge components identified by domain experts, we incorporated an automatic skill discovery method (SK), which includes a nonparametric prior over the exercise-skill assignments, to all three models. Thus, we explored a total of six models: BKT, BKT+SK, IBKT, IBKT+SK, LSTM, and LSTM+SK. Two training datasets were employed, one was collected from a natural language physics intelligent tutoring system named Cordillera, and the other was from a standard probability intelligent tutoring system named Pyrenees. Overall, our results showed that BKT and BKT+SK outperformed the others on predicting post-test scores, whereas LSTM and LSTM+SK achieved the highest accuracy, F1-measure, and area under the ROC curve (AUC) on predicting learning gains. Furthermore, we demonstrated that by combining SK with the BKT model, BKT+SK could reliably predict post-test scores using only the earliest 50% of the entire training sequences. For learning gain early prediction, using the earliest 70% of the entire sequences, LSTM can deliver a comparable prediction as using the entire training sequences. The findings yield a learning environment that can foretell students’ performance and learning gains early, and can render adaptive pedagogical strategy accordingly.
more »
« less
Deep Learning on a Data Diet: Finding Important Examples Early in Training
Recent success in deep learning has partially been driven by training increasingly overparametrized networks on ever larger datasets. It is therefore natural to ask: how much of the data is superfluous, which examples are important for general- ization, and how do we find them? In this work, we make the striking observation that, in standard vision datasets, simple scores averaged over several weight ini- tializations can be used to identify important examples very early in training. We propose two such scores—the Gradient Normed (GraNd) and the Error L2-Norm (EL2N) scores—and demonstrate their efficacy on a range of architectures and datasets by pruning significant fractions of training data without sacrificing test accuracy. In fact, using EL2N scores calculated a few epochs into training, we can prune half of the CIFAR10 training set while slightly improving test accuracy. Furthermore, for a given dataset, EL2N scores from one architecture or hyper- parameter configuration generalize to other configurations. Compared to recent work that prunes data by discarding examples that are rarely forgotten over the course of training, our scores use only local information early in training. We also use our scores to detect noisy examples and study training dynamics through the lens of important examples—we investigate how the data distribution shapes the loss surface and identify subspaces of the model’s data representation that are relatively stable over training.
more »
« less
- Award ID(s):
- 1845166
- PAR ID:
- 10513486
- Publisher / Repository:
- NeurIPS
- Date Published:
- Journal Name:
- NeurIPS
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
A variety of fairness constraints have been proposed in the literature to mitigate group-level statistical bias. Their impacts have been largely evaluated for different groups of populations corresponding to a set of sensitive attributes, such as race or gender. Nonetheless, the community has not observed sufficient explorations for how imposing fairness constraints fare at an instance level. Building on the concept of influence function, a measure that characterizes the impact of a training example on the target model and its predictive performance, this work studies the influence of training examples when fairness constraints are imposed. We find out that under certain assumptions, the influence function with respect to fairness constraints can be decomposed into a kernelized combination of training examples. One promising application of the proposed fairness influence function is to identify suspicious training examples that may cause model discrimination by ranking their influence scores. We demonstrate with extensive experiments that training on a subset of weighty data examples leads to lower fairness violations with a trade-off of accuracy.more » « less
-
Rubin, Stuart; Chen, Shu-Ching (Ed.)In this work, we use an unsupervised method for generating binary class labels in a novel context to create class labels for Medicare fraud detection. We examine how class imbalance influences the quality of these new labels and how it affects supervised classification. We use four different Medicare Part D fraud detection datasets, with the largest containing over 5 million instances. The other three datasets are sampled from the original dataset. Using Random Under-Sampling (RUS), we subsample from the majority class of the original data to produce three datasets with varying levels of class imbalance. To evaluate the performance of the newly created labels, we train a supervised classifier and evaluate its classification performance and compare it to an unsupervised anomaly detection method as a baseline. Our empirical findings indicate that the generated class labels are of high enough quality and enable effective supervised classifier training for fraud detection. Additionally, supervised classification with the new labels consistently outperforms the baseline used for comparison across all test scenarios. Further more, we observe an inverse relationship between class imbalance in the dataset and classifier performance, with AUPRC scores improving as the training dataset becomes more balanced. This work not only validates the efficacy of the synthesized class labels in labeling Medicare fraud but also shows its robustness across different degrees of class imbalance.more » « less
-
null (Ed.)Because of the lack of expertise, to gain benefits from their data, average users have to upload their private data to cloud servers they may not trust. Due to legal or privacy constraints, most users are willing to contribute only their encrypted data, and lack interests or resources to join deep neural network (DNN) training in cloud. To train a DNN on encrypted data in a completely non-interactive way, a recent work proposes a fully homomorphic encryption (FHE)-based technique implementing all activations by \textit{Brakerski-Gentry-Vaikuntanathan} (BGV)-based lookup tables. However, such inefficient lookup-table-based activations significantly prolong private training latency of DNNs. In this paper, we propose, Glyph, an FHE-based technique to fast and accurately train DNNs on encrypted data by switching between TFHE (Fast Fully Homomorphic Encryption over the Torus) and BGV cryptosystems. Glyph uses logic-operation-friendly TFHE to implement nonlinear activations, while adopts vectorial-arithmetic-friendly BGV to perform multiply-accumulations (MACs). Glyph further applies transfer learning on DNN training to improve test accuracy and reduce the number of MACs between ciphertext and ciphertext in convolutional layers. Our experimental results show Glyph obtains state-of-the-art accuracy, and reduces training latency by 69%~99% over prior FHE-based privacy-preserving techniques on encrypted datasets.more » « less
-
The Random Forests classifier, a widely utilized off-the-shelf classification tool, assumes training and test samples come from the same distribution as other standard classifiers. However, in safety-critical scenarios like medical diagnosis and network attack detection, discrepancies between the training and test sets, including the potential presence of novel outlier samples not appearing during training, can pose significant challenges. To address this problem, we introduce the Conformalized Semi-Supervised Random Forest (CSForest), which couples the conformalization technique Jackknife+aB with semi-supervised tree ensembles to construct a set-valued prediction 𝐶(𝑥). Instead of optimizing over the training distribution, CSForest employs unlabeled test samples to enhance accuracy and flag unseen outliers by generating an empty set. Theoretically, we establish CSForest to cover true labels for previously observed inlier classes under arbitrarily label-shift in the test data. We compare CSForest with state-of-the-art methods using synthetic examples and various real-world datasets, under different types of distribution changes in the test domain. Our results highlight CSForest’s effective prediction of inliers and its ability to detect outlier samples unique to the test data. In addition, CSForest shows persistently good performance as the sizes of the training and test sets vary. Codes of CSForest are available at https://github.com/yujinhan98/CSForest.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
