An official website of the United States government
Large language models (LLMs) have revolution- ized machine learning due to their ability to cap- ture complex interactions between input features. Popular post-hoc explanation methods like SHAP provide marginal feature attributions, while their extensions to interaction importances only scale to small input lengths (≈20). We propose Spectral Ex- plainer (SPEX), a model-agnostic interaction attri- bution algorithm that efficiently scales to large input lengths (≈1000). SPEX exploits underlying nat- ural sparsity among interactions—common in real- world data—and applies a sparse Fourier transform using a channel decoding algorithm to efficiently identify important interactions. We perform exper- iments across three difficult long-context datasets that require LLMs to utilize interactions between inputs to complete the task. For large inputs, SPEX outperforms marginal attribution methods by up to 20% in terms of faithfully reconstructing LLM out- puts. Further, SPEX successfully identifies key fea- tures and interactions that strongly influence model output. For one of our datasets, HotpotQA, SPEX provides interactions that align with human annota- tions. Finally, we use our model-agnostic approach to generate explanations to demonstrate abstract rea- soning in closed-source LLMs (GPT-4o mini) and compositional reasoning in vision-language models.
more »
« less
Model-agnostic Methods for Text Classification with Inherent Noise
Text classification is a fundamental problem, and recently, deep neural networks (DNN) have shown promising results in many natural language tasks. However, their human-level performance relies on high-quality annotations, which are time-consuming and expensive to collect. As we move towards large inexpensive datasets, the inherent label noise degrades the generalization of DNN. While most machine learning literature focuses on building complex networks to handle noise, in this work, we evaluate model-agnostic methods to handle inherent noise in large scale text classification that can be easily incorporated into existing machine learning workflows with minimal interruption. Specifically, we conduct a point-by-point comparative study between several noise-robust methods on three datasets encompassing three popular classification models. To our knowledge, this is the first time such a comprehensive study in text classification encircling popular models and model-agnostic loss methods has been conducted. In this study, we describe our learning and demonstrate the application of our approach, which outperformed baselines by up to 10% in classification accuracy while requiring no network modifications.
more »
« less
- Award ID(s):
- 1838159
- PAR ID:
- 10294436
- Date Published:
- Journal Name:
- 28th International Conference on Computational Linguistics: Industry Track
- Page Range / eLocation ID:
- 202–213
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The wide deployment of Deep Neural Networks (DNN) in high-performance cloud computing platforms brought to light multi-tenant cloud field-programmable gate arrays (FPGA) as a popular choice of accelerator to boost performance due to its hardware reprogramming flexibility. Such a multi-tenant FPGA setup for DNN acceleration potentially exposes DNN interference tasks under severe threat from malicious users. This work, to the best of our knowledge, is the first to explore DNN model vulnerabilities in multi-tenant FPGAs. We propose a novel adversarial attack framework: Deep-Dup, in which the adversarial tenant can inject adversarial faults to the DNN model in the victim tenant of FPGA. Specifically, she can aggressively overload the shared power distribution system of FPGA with malicious power-plundering circuits, achieving adversarial weight duplication (AWD) hardware attack that duplicates certain DNN weight packages during data transmission between off-chip memory and on-chip buffer, to hijack the DNN function of the victim tenant. Further, to identify the most vulnerable DNN weight packages for a given malicious objective, we propose a generic vulnerable weight package searching algorithm, called Progressive Differential Evolution Search (P-DES), which is, for the first time, adaptive to both deep learning white-box and black-box attack models. The proposed Deep-Dup is experimentally validated in a developed multi-tenant FPGA prototype, for two popular deep learning applications, i.e., Object Detection and Image Classification. Successful attacks are demonstrated in six popular DNN architectures (e.g., YOLOv2, ResNet-50, MobileNet, etc.) on three datasets (COCO, CIFAR-10, and ImageNet).more » « less
-
Effectively filtering and categorizing the large volume of user-generated content on social media during disaster events can help emergency management and disaster response prioritize their resources. Deep learning approaches, including recurrent neural networks and transformer-based models, have been previously used for this purpose. Capsule Neural Networks (CapsNets), initially proposed for image classification, have been proven to be useful for text analysis as well. However, to the best of our knowledge, CapsNets have not been used for classifying crisis-related messages, and have not been extensively compared with state-of-the-art transformer-based models, such as BERT. Therefore, in this study, we performed a thorough comparison between CapsNet models, state-of-the-art BERT models and two popular recurrent neural network models that have been successfully used for tweet classification, specifically, LSTM and Bi-LSTM models, on the task of classifying crisis tweets both in terms of their informativeness (binary classification), as well as their humanitarian content (multi-class classification). For this purpose, we used several benchmark datasets for crisis tweet classification, namely CrisisBench, CrisisNLP and CrisisLex. Experimental results show that the performance of the CapsNet models is on a par with that of LSTM and Bi-LSTM models for all metrics considered, while the performance obtained with BERT models have surpassed the performance of the other three models across different datasets and classes for both classification tasks, and thus BERT could be considered the best overall model for classifying crisis tweets.more » « less
-
Deep neural networks are gaining increasing popularity for the classic text classification task, due to their strong expressive power and less requirement for feature engineering. Despite such attractiveness, neural text classification models suffer from the lack of training data in many real-world applications. Although many semisupervised and weakly-supervised text classification models exist, they cannot be easily applied to deep neural models and meanwhile support limited supervision types. In this paper, we propose a weakly-supervised method that addresses the lack of training data in neural text classification. Our method consists of two modules: (1) a pseudo-document generator that leverages seed information to generate pseudo-labeled documents for model pre-training, and (2) a self-training module that bootstraps on real unlabeled data for model refinement. Our method has the flexibility to handle different types of weak supervision and can be easily integrated into existing deep neural models for text classification. We have performed extensive experiments on three real-world datasets from different domains. The results demonstrate that our proposed method achieves inspiring performance without requiring excessive training data and outperforms baseline methods significantly.more » « less
-
Network intrusion detection systems (NIDS) today must quickly provide visibility into anomalous behavior on a growing amount of data. Meanwhile different data models have evolved over time, each providing a different set of features to classify attacks. Defenders have limited time to retrain classifiers, while the scale of data and feature mismatch between data models can affect the ability to periodically retrain. Much work has focused on classification accuracy yet feature selection is a key part of machine learning that, when optimized, reduces the training time and can increase accuracy by removing poorly performing features that introduce noise. With a larger feature space, the pursuit of more features is not as valuable as selecting better features. In this paper, we use an ensemble approach of filter methods to rank features followed by a voting technique to select a subset of features. We evaluate our approach using three datasets to show that, across datasets and network topologies, similar features have a trivial effect on classifier accuracy after removal. Our approach identifies poorly performing features to remove in a classifier-agnostic manner that can significantly save time for periodic retraining of production NIDS.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.18653/v1/2020.coling-industry.19
