More Like this

1. Keeping up with the genomes: efficient learning of our increasing knowledge of the tree of life

   https://doi.org/10.1186/s12859-020-03744-7  

   Zhao, Zhengqiao; Cristian, Alexandru; Rosen, Gail (December 2020, BMC Bioinformatics)

   null (Ed.)

   Abstract Background It is a computational challenge for current metagenomic classifiers to keep up with the pace of training data generated from genome sequencing projects, such as the exponentially-growing NCBI RefSeq bacterial genome database. When new reference sequences are added to training data, statically trained classifiers must be rerun on all data, resulting in a highly inefficient process. The rich literature of “incremental learning” addresses the need to update an existing classifier to accommodate new data without sacrificing much accuracy compared to retraining the classifier with all data. Results We demonstrate how classification improves over time by incrementally training a classifier on progressive RefSeq snapshots and testing it on: (a) all known current genomes (as a ground truth set) and (b) a real experimental metagenomic gut sample. We demonstrate that as a classifier model’s knowledge of genomes grows, classification accuracy increases. The proof-of-concept naïve Bayes implementation, when updated yearly, now runs in 1/4 t h of the non-incremental time with no accuracy loss. Conclusions It is evident that classification improves by having the most current knowledge at its disposal. Therefore, it is of utmost importance to make classifiers computationally tractable to keep up with the data deluge. The incremental learning classifier can be efficiently updated without the cost of reprocessing nor the access to the existing database and therefore save storage as well as computation resources. 

   more » « less
2. Self-Stabilization: The Implicit Bias of Gradient Descent at the Edge of Stability.

   Damian, Alex; Nichani, Eshaan; Lee, Jason D. (May 2023, ICLR 2023)

   Traditional analyses of gradient descent show that when the largest eigenvalue of the Hessian, also known as the sharpness S(θ), is bounded by 2/η, training is "stable" and the training loss decreases monotonically. Recent works, however, have observed that this assumption does not hold when training modern neural networks with full batch or large batch gradient descent. Most recently, Cohen et al. (2021) observed two important phenomena. The first, dubbed progressive sharpening, is that the sharpness steadily increases throughout training until it reaches the instability cutoff 2/η. The second, dubbed edge of stability, is that the sharpness hovers at 2/η for the remainder of training while the loss continues decreasing, albeit non-monotonically. We demonstrate that, far from being chaotic, the dynamics of gradient descent at the edge of stability can be captured by a cubic Taylor expansion: as the iterates diverge in direction of the top eigenvector of the Hessian due to instability, the cubic term in the local Taylor expansion of the loss function causes the curvature to decrease until stability is restored. This property, which we call self-stabilization, is a general property of gradient descent and explains its behavior at the edge of stability. A key consequence of self-stabilization is that gradient descent at the edge of stability implicitly follows projected gradient descent (PGD) under the constraint S(θ)≤2/η. Our analysis provides precise predictions for the loss, sharpness, and deviation from the PGD trajectory throughout training, which we verify both empirically in a number of standard settings and theoretically under mild conditions. Our analysis uncovers the mechanism for gradient descent's implicit bias towards stability. 

   more » « less
3. Membership Inference Attacks and Defenses in Classification Models

   Li, Jiacheng and (July 2021, Proceedings of the Eleventh ACM Conference on Data and Application Security and Privacy)

   We study the membership inference (MI) attack against classifiers, where the attacker's goal is to determine whether a data instance was used for training the classifier. Through systematic cataloging of existing MI attacks and extensive experimental evaluations of them, we find that a model's vulnerability to MI attacks is tightly related to the generalization gap -- the difference between training accuracy and test accuracy. We then propose a defense against MI attacks that aims to close the gap by intentionally reduces the training accuracy. More specifically, the training process attempts to match the training and validation accuracies, by means of a new {\em set regularizer} using the Maximum Mean Discrepancy between the softmax output empirical distributions of the training and validation sets. Our experimental results show that combining this approach with another simple defense (mix-up training) significantly improves state-of-the-art defense against MI attacks, with minimal impact on testing accuracy. 

   more » « less
4. Data Mining with Algorithmic Transparency

   https://doi.org/10.1007/978-3-319-93034-3_11  

   Yan Zhou, Yasmeen Alufaisan (January 2018, PAKDD 2018: Advances in Knowledge Discovery and Data Mining)

   In this paper, we investigate whether decision trees can be used to interpret a black-box classifier without knowing the learning algorithm and the training data. Decision trees are known for their transparency and high expressivity. However, they are also notorious for their instability and tendency to grow excessively large. We present a classifier reverse engineering model that outputs a decision tree to interpret the black-box classifier. There are two major challenges. One is to build such a decision tree with controlled stability and size, and the other is that probing the black-box classifier is limited for security and economic reasons. Our model addresses the two issues by simultaneously minimizing sampling cost and classifier complexity. We present our empirical results on four real datasets, and demonstrate that our reverse engineering learning model can effectively approximate and simplify the black box classifier. 

   more » « less
5. Global Multiclass Classification and Dataset Construction via Heterogeneous Local Experts

   https://doi.org/10.1109/JSAIT.2020.3041804  

   Ahn, Surin; Ozgur, Ayfer; Pilanci, Mert (January 2021, IEEE journal on selected areas in information theory)

   In the domains of dataset construction and crowdsourcing, a notable challenge is to aggregate labels from a heterogeneous set of labelers, each of whom is potentially an expert in some subset of tasks (and less reliable in others). To reduce costs of hiring human labelers or training automated labeling systems, it is of interest to minimize the number of labelers while ensuring the reliability of the resulting dataset. We model this as the problem of performing K-class classification using the predictions of smaller classifiers, each trained on a subset of [K], and derive bounds on the number of classifiers needed to accurately infer the true class of an unlabeled sample under both adversarial and stochastic assumptions. By exploiting a connection to the classical set cover problem, we produce a near-optimal scheme for designing such configurations of classifiers which recovers the well known one-vs.-one classification approach as a special case. Experiments with the MNIST and CIFAR-10 datasets demonstrate the favorable accuracy (compared to a centralized classifier) of our aggregation scheme applied to classifiers trained on subsets of the data. These results suggest a new way to automatically label data or adapt an existing set of local classifiers to larger-scale multiclass problems. 

   more » « less