Recently, it was shown that the role of the teacher in knowledge distillation (KD) is to provide the student with an estimate of the true Bayes conditional probability density (BCPD). Notably, the new findings propose that the student's error rate can be upper-bounded by the mean squared error (MSE) between the teacher's output and BCPD. Consequently, to enhance KD efficacy, the teacher should be trained such that its output is close to BCPD in MSE sense. This paper elucidates that training the teacher model with MSE loss equates to minimizing the MSE between its output and BCPD, aligning with its core responsibility of providing the student with a BCPD estimate closely resembling it in MSE terms. In this respect, through a comprehensive set of experiments, we demonstrate that substituting the conventional teacher trained with cross-entropy loss with one trained using MSE loss in state-of-the-art KD methods consistently boosts the student's accuracy, resulting in improvements of up to 2.6\%.

Language Models (LMs) have demonstrated exceptional performance across various Natural Language Processing (NLP) tasks. Despite these advancements, LMs can inherit and amplify societal biases related to sensitive attributes such as gender and race, limiting their adoption in real-world applications. Therefore, fairness has been extensively explored in LMs, leading to the proposal of various fairness notions. However, the lack of clear agreement on which fairness definition to apply in specific contexts (\textit{e.g.,} medium-sized LMs versus large-sized LMs) and the complexity of understanding the distinctions between these definitions can create confusion and impede further progress. To this end, this paper proposes a systematic survey that clarifies the definitions of fairness as they apply to LMs. Specifically, we begin with a brief introduction to LMs and fairness in LMs, followed by a comprehensive, up-to-date overview of existing fairness notions in LMs and the introduction of a novel taxonomy that categorizes these concepts based on their foundational principles and operational distinctions. We further illustrate each definition through experiments, showcasing their practical implications and outcomes. Finally, we discuss current research challenges and open questions, aiming to foster innovative ideas and advance the field. The implementation and additional resources are publicly available at https://github.com/LavinWong/Fairness-in-Large-Language-Models/tree/main/definitions.

Learning models whose predictions are invariant under multiple environments is a promising approach for out-of-distribution generalization. Such models are trained to extract features $X_{\text{inv}}$ where the conditional distribution $Y \mid X_{\text{inv}}$ of the label given the extracted features does not change across environments. Invariant models are also supposed to generalize to shifts in the marginal distribution $p(X_{\text{inv}})$ of the extracted features $X_{\text{inv}}$, a type of shift we call an $\textit{invariant covariate shift}$. However, we show that proposed methods for learning invariant models underperform under invariant covariate shift, either failing to learn invariant models$\unicode{x2014}$even for data generated from simple and well-studied linear-Gaussian models$\unicode{x2014}$or having poor finite-sample performance. To alleviate these problems, we propose $\textit{weighted risk invariance}$ (WRI). Our framework is based on imposing invariance of the loss across environments subject to appropriate reweightings of the training examples. We show that WRI provably learns invariant models, i.e. discards spurious correlations, in linear-Gaussian settings. We propose a practical algorithm to implement WRI by learning the density $p(X_{\text{inv}})$ and the model parameters simultaneously, and we demonstrate empirically that WRI outperforms previous invariant learning methods under invariant covariate shift.

Rapid development of disease detection models using computer vision is crucial in responding to medical emergencies, such as epidemics or bioterrorism events. Traditional data collection methods are often too slow in these scenarios, requiring innovative approaches for quick, reliable model generation from minimal data. Our study introduces a novel approach by constructing a comprehensive computer vision model to detect Mpox lesions using only synthetic data. Initially, these models generated a diverse set of synthetic images representing Mpox lesions on various body parts (face, back, chest, leg, neck, arm) across different skin tones as defined by the Fitzpatrick scale (fair, brown, dark skin). Subsequently, we trained and tested a vision model with this synthetic dataset to evaluate the diffusion models' efficacy in producing high-quality training data and its impact on the vision model's medical image recognition performance. The results were promising; the vision model achieved a 97% accuracy rate, with 96% precision and recall for Mpox cases, and similarly high metrics for normal and other skin disorder cases, demonstrating its ability to correctly identify true positives and minimize false positives. The model achieved an F1-Score of 96% for Mpox cases and 98% for normal and other skin disorders, reflecting a balanced precision-recall relationship, thus ensuring reliability and robustness in its predictions. Our proposed SynthVision methodology indicates the potential to develop accurate computer vision models with minimal data input for future medical emergencies.

In this study, state-of-the-art unsupervised detection models were evaluated for the purpose of automated anomaly inspection of wool carpets. A custom dataset of four unique types of carpet textures was created to thoroughly test the models and their robustness in detecting subtle anomalies in complex textures. Due to the requirements of an inline inspection system in a manufacturing use case, the metrics of importance in this study were accuracy in detecting anomalous areas, the number of false detections, and the inference times of each model for real-time performance. Of the evaluated models, the student-teacher network based methods were found on average to yield the highest detection accuracy and lowest false detection rates. When trained on a multi-class dataset the models were found to yield comparable if not better results than single-class training. Finally, in terms of detection speed, with exception to the generative model, all other evaluated models were found to have comparable inference times on a GPU, with an average of 0.16s per image. On a CPU, most of these models typically produced results between 1.5 to 2 times the respective GPU inference times.

We investigate whether an LLM can successfully perform financial statement analysis in a way similar to a professional human analyst. We provide standardized and anonymous financial statements to GPT4 and instruct the model to analyze them to determine the direction of future earnings. Even without any narrative or industry-specific information, the LLM outperforms financial analysts in its ability to predict earnings changes. The LLM exhibits a relative advantage over human analysts in situations when the analysts tend to struggle. Furthermore, we find that the prediction accuracy of the LLM is on par with the performance of a narrowly trained state-of-the-art ML model. LLM prediction does not stem from its training memory. Instead, we find that the LLM generates useful narrative insights about a company's future performance. Lastly, our trading strategies based on GPT's predictions yield a higher Sharpe ratio and alphas than strategies based on other models. Taken together, our results suggest that LLMs may take a central role in decision-making.

Background: Software Vulnerability (SV) prediction needs large-sized and high-quality data to perform well. Current SV datasets mostly require expensive labeling efforts by experts (human-labeled) and thus are limited in size. Meanwhile, there are growing efforts in automatic SV labeling at scale. However, the fitness of auto-labeled data for SV prediction is still largely unknown. Aims: We quantitatively and qualitatively study the quality and use of the state-of-the-art auto-labeled SV data, D2A, for SV prediction. Method: Using multiple sources and manual validation, we curate clean SV data from human-labeled SV-fixing commits in two well-known projects for investigating the auto-labeled counterparts. Results: We discover that 50+% of the auto-labeled SVs are noisy (incorrectly labeled), and they hardly overlap with the publicly reported ones. Yet, SV prediction models utilizing the noisy auto-labeled SVs can perform up to 22% and 90% better in Matthews Correlation Coefficient and Recall, respectively, than the original models. We also reveal the promises and difficulties of applying noise-reduction methods for automatically addressing the noise in auto-labeled SV data to maximize the data utilization for SV prediction. Conclusions: Our study informs the benefits and challenges of using auto-labeled SVs, paving the way for large-scale SV prediction.

This article puts forward the use of mutual information values to replicate the expertise of security professionals in selecting features for detecting web attacks. The goal is to enhance the effectiveness of web application firewalls (WAFs). Web applications are frequently vulnerable to various security threats, making WAFs essential for their protection. WAFs analyze HTTP traffic using rule-based approaches to identify known attack patterns and to detect and block potential malicious requests. However, a major challenge is the occurrence of false positives, which can lead to blocking legitimate traffic and impact the normal functioning of the application. The problem is addressed as an approach that combines supervised learning for feature selection with a semi-supervised learning scenario for training a One-Class SVM model. The experimental findings show that the model trained with features selected by the proposed algorithm outperformed the expert-based selection approach in terms of performance. Additionally, the results obtained by the traditional rule-based WAF ModSecurity, configured with a vanilla set of OWASP CRS rules, were also improved.

Vision-based inspection algorithms have significantly contributed to quality control in industrial settings, particularly in addressing structural defects like dent and contamination which are prevalent in mass production. Extensive research efforts have led to the development of related benchmarks such as MVTec AD (Bergmann et al., 2019). However, in industrial settings, there can be instances of logical defects, where acceptable items are found in unsuitable locations or product pairs do not match as expected. Recent methods tackling logical defects effectively employ knowledge distillation to generate difference maps. Knowledge distillation (KD) is used to learn normal data distribution in unsupervised manner. Despite their effectiveness, these methods often overlook the potential false negatives. Excessive similarity between the teacher network and student network can hinder the generation of a suitable difference map for logical anomaly detection. This technical report provides insights on handling potential false negatives by utilizing a simple constraint in KD-based logical anomaly detection methods. We select EfficientAD as a state-of-the-art baseline and apply a margin-based constraint to its unsupervised learning scheme. Applying this constraint, we can improve the AUROC for MVTec LOCO AD by 1.3 %.

In this paper, we develop an unsupervised learning approach to earthquake detection. We train a specific class of deep auto-encoders that learn to reproduce the input waveforms after a data-compressive bottleneck, and then use a simple triggering algorithm at the bottleneck to label waveforms as noise or signal. Our approach is motivated by the intuition that efficient compression of data should represent signals differently from noise, and is facilitated by a time-axis-preserving approach to auto-encoding and intuitively-motivated choices on the architecture and triggering. We demonstrate that the detection performance of the unsupervised approach is comparable to, and in some cases better than, some of the state-of-the-art supervised methods. Moreover, it has strong \emph{cross-dataset generalization}. By experimenting with various modifications, we demonstrate that the detection performance is insensitive to various technical choices made in the algorithm. Our approach has the potential to be useful for other signal detection problems with time series data.