Machine learning methods can detect Android malware with very high accuracy. However, these classifiers have an Achilles heel, concept drift: they rapidly become out of date and ineffective, due to the evolution of malware apps and benign apps. Our research finds that, after training an Android malware classifier on one year's worth of data, the F1 score quickly dropped from 0.99 to 0.76 after 6 months of deployment on new test samples. In this paper, we propose new methods to combat the concept drift problem of Android malware classifiers. Since machine learning technique needs to be continuously deployed, we use active learning: we select new samples for analysts to label, and then add the labeled samples to the training set to retrain the classifier. Our key idea is, similarity-based uncertainty is more robust against concept drift. Therefore, we combine contrastive learning with active learning. We propose a new hierarchical contrastive learning scheme, and a new sample selection technique to continuously train the Android malware classifier. Our evaluation shows that this leads to significant improvements, compared to previously published methods for active learning. Our approach reduces the false negative rate from 14% (for the best baseline) to 9%, while also reducing the false positive rate (from 0.86% to 0.48%). Also, our approach maintains more consistent performance across a seven-year time period than past methods.

Given a user's input text, text-matching recommender systems output relevant items by comparing the input text to available items' description, such as product-to-product recommendation on e-commerce platforms. As users' interests and item inventory are expected to change, it is important for a text-matching system to generalize to data shifts, a task known as out-of-distribution (OOD) generalization. However, we find that the popular approach of fine-tuning a large, base language model on paired item relevance data (e.g., user clicks) can be counter-productive for OOD generalization. For a product recommendation task, fine-tuning obtains worse accuracy than the base model when recommending items in a new category or for a future time period. To explain this generalization failure, we consider an intervention-based importance metric, which shows that a fine-tuned model captures spurious correlations and fails to learn the causal features that determine the relevance between any two text inputs. Moreover, standard methods for causal regularization do not apply in this setting, because unlike in images, there exist no universally spurious features in a text-matching task (the same token may be spurious or causal depending on the text it is being matched to). For OOD generalization on text inputs, therefore, we highlight a different goal: avoiding high importance scores for certain features. We do so using an intervention-based regularizer that constraints the causal effect of any token on the model's relevance score to be similar to the base model. Results on Amazon product and 3 question recommendation datasets show that our proposed regularizer improves generalization for both in-distribution and OOD evaluation, especially in difficult scenarios when the base model is not accurate.

Sepsis is a life-threatening condition with organ dysfunction and is a leading cause of death and critical illness worldwide. Even a few hours of delay in the treatment of sepsis results in increased mortality. Early detection of sepsis during emergency department triage would allow early initiation of lab analysis, antibiotic administration, and other sepsis treatment protocols. The purpose of this study was to compare sepsis detection performance at ED triage (prior to the use of laboratory diagnostics) of the standard sepsis screening algorithm (SIRS with source of infection) and a machine learning algorithm trained on EHR triage data. A machine learning model (KATE Sepsis) was developed using patient encounters with triage data from 16participating hospitals. KATE Sepsis and standard screening were retrospectively evaluated on the adult population of 512,949 medical records. KATE Sepsis demonstrates an AUC of 0.9423 (0.9401 - 0.9441) with sensitivity of 71.09% (70.12% - 71.98%) and specificity of 94.81% (94.75% - 94.87%). Standard screening demonstrates an AUC of 0.6826 (0.6774 - 0.6878) with sensitivity of 40.8% (39.71% - 41.86%) and specificity of 95.72% (95.68% - 95.78%). The KATE Sepsis model trained to detect sepsis demonstrates 77.67% (75.78% -79.42%) sensitivity in detecting severe sepsis and 86.95% (84.2% - 88.81%) sensitivity in detecting septic shock. The standard screening protocol demonstrates 43.06% (41% - 45.87%) sensitivity in detecting severe sepsis and40% (36.55% - 43.26%) sensitivity in detecting septic shock. Future research should focus on the prospective impact of KATE Sepsis on administration of antibiotics, readmission rate, morbidity and mortality.

Suppose we are given two datasets: a labeled dataset and unlabeled dataset which also has additional auxiliary features not present in the first dataset. What is the most principled way to use these datasets together to construct a predictor? The answer should depend upon whether these datasets are generated by the same or different distributions over their mutual feature sets, and how similar the test distribution will be to either of those distributions. In many applications, the two datasets will likely follow different distributions, but both may be close to the test distribution. We introduce the problem of building a predictor which minimizes the maximum loss over all probability distributions over the original features, auxiliary features, and binary labels, whose Wasserstein distance is $r_1$ away from the empirical distribution over the labeled dataset and $r_2$ away from that of the unlabeled dataset. This can be thought of as a generalization of distributionally robust optimization (DRO), which allows for two data sources, one of which is unlabeled and may contain auxiliary features.

Training machine learning models on privacy-sensitive data has become a popular practice, driving innovation in ever-expanding fields. This has opened the door to new attacks that can have serious privacy implications. One such attack, the Membership Inference Attack (MIA), exposes whether or not a particular data point was used to train a model. A growing body of literature uses Differentially Private (DP) training algorithms as a defence against such attacks. However, these works evaluate the defence under the restrictive assumption that all members of the training set, as well as non-members, are independent and identically distributed. This assumption does not hold for many real-world use cases in the literature. Motivated by this, we evaluate membership inference with statistical dependencies among samples and explain why DP does not provide meaningful protection (the privacy parameter $\epsilon$ scales with the training set size $n$) in this more general case. We conduct a series of empirical evaluations with off-the-shelf MIAs using training sets built from real-world data showing different types of dependencies among samples. Our results reveal that training set dependencies can severely increase the performance of MIAs, and therefore assuming that data samples are statistically independent can significantly underestimate the performance of MIAs.

Hotel bathrooms are one of the most important places in terms of customer satisfaction, and where the most complaints are reported. To share their experiences, guests rate hotels, comment, and share images of their positive or negative ratings. An important part of the room images shared by guests is related to bathrooms. Guests tend to prove their satisfaction or dissatisfaction with the bathrooms with images in their comments. These Positive or negative comments and visuals potentially affect the prospective guests. In this study, two different versions of a deep learning algorithm were designed to classify hotel bathrooms as satisfactory (good) or unsatisfactory (bad, when any defects such as dirtiness, deficiencies, malfunctions were present) by analyzing images. The best-performer between the two models was determined as a result of a series of extensive experimental studies. The models were trained for each of 144 combinations of 5 hyper-parameter sets with a data set containing more than 11 thousand bathroom images, specially created for this study. The "HotelBath" data set was shared also with the community with this study. Four different image sizes were taken into consideration: 128, 256, 512 and 1024 pixels in both directions. The classification performances of the models were measured with several metrics. Both algorithms showed very attractive performances even with many combinations of hyper-parameters. They can classify bathroom images with very high accuracy. Suh that the top algorithm achieved an accuracy of 92.4% and an AUC (area under the curve) score of 0.967. In addition, other metrics also proved the success...

The increasing reliance on large language models (LLMs) in academic writing has led to a rise in plagiarism. Existing AI-generated text classifiers have limited accuracy and often produce false positives. We propose a novel approach using natural language processing (NLP) techniques, offering quantifiable metrics at both sentence and document levels for easier interpretation by human evaluators. Our method employs a multi-faceted approach, generating multiple paraphrased versions of a given question and inputting them into the LLM to generate answers. By using a contrastive loss function based on cosine similarity, we match generated sentences with those from the student's response. Our approach achieves up to 94% accuracy in classifying human and AI text, providing a robust and adaptable solution for plagiarism detection in academic settings. This method improves with LLM advancements, reducing the need for new model training or reconfiguration, and offers a more transparent way of evaluating and detecting AI-generated text.

As the size of the pre-trained language model (PLM) continues to increase, numerous parameter-efficient transfer learning methods have been proposed recently to compensate for the tremendous cost of fine-tuning. Despite the impressive results achieved by large pre-trained language models (PLMs) and various parameter-efficient transfer learning (PETL) methods on sundry benchmarks, it remains unclear if they can handle inputs that have been distributionally shifted effectively. In this study, we systematically explore how the ability to detect out-of-distribution (OOD) changes as the size of the PLM grows or the transfer methods are altered. Specifically, we evaluated various PETL techniques, including fine-tuning, Adapter, LoRA, and prefix-tuning, on three different intention classification tasks, each utilizing various language models with different scales.

Large language models (LLMs) are remarkable data annotators. They can be used to generate high-fidelity supervised training data, as well as survey and experimental data. With the widespread adoption of LLMs, human gold--standard annotations are key to understanding the capabilities of LLMs and the validity of their results. However, crowdsourcing, an important, inexpensive way to obtain human annotations, may itself be impacted by LLMs, as crowd workers have financial incentives to use LLMs to increase their productivity and income. To investigate this concern, we conducted a case study on the prevalence of LLM usage by crowd workers. We reran an abstract summarization task from the literature on Amazon Mechanical Turk and, through a combination of keystroke detection and synthetic text classification, estimate that 33-46% of crowd workers used LLMs when completing the task. Although generalization to other, less LLM-friendly tasks is unclear, our results call for platforms, researchers, and crowd workers to find new ways to ensure that human data remain human, perhaps using the methodology proposed here as a stepping stone. Code/data: https://github.com/epfl-dlab/GPTurk

In this paper, we develop a series of differential privacy (DP) algorithms from a family of random projections (RP) for general applications in machine learning, data mining, and information retrieval. Among the presented algorithms, iDP-SignRP is remarkably effective under the setting of ``individual differential privacy'' (iDP), based on sign random projections (SignRP). Also, DP-SignOPORP considerably improves existing algorithms in the literature under the standard DP setting, using ``one permutation + one random projection'' (OPORP), where OPORP is a variant of the celebrated count-sketch method with fixed-length binning and normalization. Without taking signs, among the DP-RP family, DP-OPORP achieves the best performance. Our key idea for improving DP-RP is to take only the signs, i.e., $sign(x_j) = sign\left(\sum_{i=1}^p u_i w_{ij}\right)$, of the projected data. The intuition is that the signs often remain unchanged when the original data ($u$) exhibit small changes (according to the ``neighbor'' definition in DP). In other words, the aggregation and quantization operations themselves provide good privacy protections. We develop a technique called ``smooth flipping probability'' that incorporates this intuitive privacy benefit of SignRPs and improves the standard DP bit flipping strategy. Based on this technique, we propose DP-SignOPORP which satisfies strict DP and outperforms other DP variants based on SignRP (and RP), especially when $\epsilon$ is not very large (e.g., $\epsilon = 5\sim10$). Moreover, if an application scenario accepts individual DP, then we immediately obtain an algorithm named iDP-SignRP which achieves excellent utilities even at small~$\epsilon$ (e.g., $\epsilon<0.5$).