Self-supervised learning in computer vision trains on unlabeled data, such as images or (image, text) pairs, to obtain an image encoder that learns high-quality embeddings for input data. Emerging backdoor attacks towards encoders expose crucial vulnerabilities of self-supervised learning, since downstream classifiers (even further trained on clean data) may inherit backdoor behaviors from encoders. Existing backdoor detection methods mainly focus on supervised learning settings and cannot handle pre-trained encoders especially when input labels are not available. In this paper, we propose DECREE, the first backdoor detection approach for pre-trained encoders, requiring neither classifier headers nor input labels. We evaluate DECREE on over 400 encoders trojaned under 3 paradigms. We show the effectiveness of our method on image encoders pre-trained on ImageNet and OpenAI's CLIP 400 million image-text pairs. Our method consistently has a high detection accuracy even if we have only limited or no access to the pre-training dataset.

Today, many systems use artificial intelligence (AI) to solve complex problems. While this often increases system effectiveness, developing a production-ready AI-based system is a difficult task. Thus, solid AI engineering practices are required to ensure the quality of the resulting system and to improve the development process. While several practices have already been proposed for the development of AI-based systems, detailed practical experiences of applying these practices are rare. In this paper, we aim to address this gap by collecting such experiences during a case study, namely the development of an autonomous stock trading system that uses machine learning functionality to invest in stocks. We selected 10 AI engineering practices from the literature and systematically applied them during development, with the goal to collect evidence about their applicability and effectiveness. Using structured field notes, we documented our experiences. Furthermore, we also used field notes to document challenges that occurred during the development, and the solutions we applied to overcome them. Afterwards, we analyzed the collected field notes, and evaluated how each practice improved the development. Lastly, we compared our evidence with existing literature. Most applied practices improved our system, albeit to varying extent, and we were able to overcome all major challenges. The qualitative results provide detailed accounts about 10 AI engineering practices, as well as challenges and solutions associated with such a project. Our experiences therefore enrich the emerging body of evidence in this field, which may be especially helpful for practitioner teams new to AI engineering.

Molecular classification has transformed the management of brain tumors by enabling more accurate prognostication and personalized treatment. However, timely molecular diagnostic testing for patients with brain tumors is limited, complicating surgical and adjuvant treatment and obstructing clinical trial enrollment. In this study, we developed DeepGlioma, a rapid ($< 90$ seconds), artificial-intelligence-based diagnostic screening system to streamline the molecular diagnosis of diffuse gliomas. DeepGlioma is trained using a multimodal dataset that includes stimulated Raman histology (SRH); a rapid, label-free, non-consumptive, optical imaging method; and large-scale, public genomic data. In a prospective, multicenter, international testing cohort of patients with diffuse glioma ($n=153$) who underwent real-time SRH imaging, we demonstrate that DeepGlioma can predict the molecular alterations used by the World Health Organization to define the adult-type diffuse glioma taxonomy (IDH mutation, 1p19q co-deletion and ATRX mutation), achieving a mean molecular classification accuracy of $93.3\pm 1.6\%$. Our results represent how artificial intelligence and optical histology can be used to provide a rapid and scalable adjunct to wet lab methods for the molecular screening of patients with diffuse glioma.

Automatic language identification is a challenging problem. Discriminating between closely related languages is especially difficult. This paper presents a machine learning approach for automatic language identification for the Nordic languages, which often suffer miscategorisation by existing state-of-the-art tools. Concretely we will focus on discrimination between six Nordic languages: Danish, Swedish, Norwegian (Nynorsk), Norwegian (Bokm{\aa}l), Faroese and Icelandic.

To illustrate everything, let us use a toy dataset with two real features x₁, x₂, and three classes c₁, c₂, c₃ in the following. Let us start with the class probability p(c), the probability that some class c is observed in the labeled dataset. The simplest way to estimate this is to just compute the relative frequencies of the classes and use them as the probabilities. We can use our dataset to see what this means exactly. There are 7 out of 20 points labeled class c₁ (blue) in the dataset, therefore we say p(c₁) 7/20.

Access to complete data in large-scale networks is often infeasible. Therefore, the problem of missing data is a crucial and unavoidable issue in the analysis and modeling of real-world social networks. However, most of the research on different aspects of social networks does not consider this limitation. One effective way to solve this problem is to recover the missing data as a pre-processing step. In this paper, a model is learned from partially observed data to infer unobserved diffusion and structure networks. To jointly discover omitted diffusion activities and hidden network structures, we develop a probabilistic generative model called "DiffStru." The interrelations among links of nodes and cascade processes are utilized in the proposed method via learning coupled with low-dimensional latent factors. Besides inferring unseen data, latent factors such as community detection may also aid in network classification problems. We tested different missing data scenarios on simulated independent cascades over LFR networks and real datasets, including Twitter and Memtracker. Experiments on these synthetic and real-world datasets show that the proposed method successfully detects invisible social behaviors, predicts links, and identifies latent features.

Diffusion models have attracted attention in recent years as innovative generative models. In this paper, we investigate whether a diffusion model is resistant to a membership inference attack, which evaluates the privacy leakage of a machine learning model. We primarily discuss the diffusion model from the standpoints of comparison with a generative adversarial network (GAN) as conventional models and hyperparameters unique to the diffusion model, i.e., time steps, sampling steps, and sampling variances. We conduct extensive experiments with DDIM as a diffusion model and DCGAN as a GAN on the CelebA and CIFAR-10 datasets in both white-box and black-box settings and then confirm if the diffusion model is comparably resistant to a membership inference attack as GAN. Next, we demonstrate that the impact of time steps is significant and intermediate steps in a noise schedule are the most vulnerable to the attack. We also found two key insights through further analysis. First, we identify that DDIM is vulnerable to the attack for small sample sizes instead of achieving a lower FID. Second, sampling steps in hyperparameters are important for resistance to the attack, whereas the impact of sampling variances is quite limited.

When an adversary provides poison samples to a machine learning model, privacy leakage, such as membership inference attacks that infer whether a sample was included in the training of the model, becomes effective by moving the sample to an outlier. However, the attacks can be detected because inference accuracy deteriorates due to poison samples. In this paper, we discuss a \textit{backdoor-assisted membership inference attack}, a novel membership inference attack based on backdoors that return the adversary's expected output for a triggered sample. We found three crucial insights through experiments with an academic benchmark dataset. We first demonstrate that the backdoor-assisted membership inference attack is unsuccessful. Second, when we analyzed loss distributions to understand the reason for the unsuccessful results, we found that backdoors cannot separate loss distributions of training and non-training samples. In other words, backdoors cannot affect the distribution of clean samples. Third, we also show that poison and triggered samples activate neurons of different distributions. Specifically, backdoors make any clean sample an inlier, contrary to poisoning samples. As a result, we confirm that backdoors cannot assist membership inference.

Artificial intelligence methods show great promise in increasing the quality and speed of work with large astronomical datasets, but the high complexity of these methods leads to the extraction of dataset-specific, nonrobust features. Therefore, such methods do not generalize well across multiple datasets. We present a universal domain adaptation method, DeepAstroUDA, as an approach to overcome this challenge. This algorithm performs semi-supervised domain adaptation and can be applied to datasets with different data distributions and class overlaps. Non-overlapping classes can be present in any of the two datasets (the labeled source domain, or the unlabeled target domain), and the method can even be used in the presence of unknown classes. We apply our method to three examples of galaxy morphology classification tasks of different complexities (3-class and 10-class problems), with anomaly detection: 1) datasets created after different numbers of observing years from a single survey (LSST mock data of 1 and 10 years of observations); 2) data from different surveys (SDSS and DECaLS); and 3) data from observing fields with different depths within one survey (wide field and Stripe 82 deep field of SDSS). For the first time, we demonstrate the successful use of domain adaptation between very discrepant observational datasets. DeepAstroUDA is capable of bridging the gap between two astronomical surveys, increasing classification accuracy in both domains (up to 40% on the unlabeled data), and making model performance consistent across datasets. Furthermore, our method also performs well as an anomaly detection algorithm and successfully clusters unknown class samples even in the unlabeled target dataset.

Undiagnosed diabetes is present in 21.4% of adults with diabetes. Diabetes can remain asymptomatic and undetected due to limitations in screening rates. To address this issue, questionnaires, such as the American Diabetes Association (ADA) Risk test, have been recommended for use by physicians and the public. Based on evidence that blood glucose concentration can affect cardiac electrophysiology, we hypothesized that an artificial intelligence (AI)-enhanced electrocardiogram (ECG) could identify adults with new-onset diabetes. We trained a neural network to estimate HbA1c using a 12-lead ECG and readily available demographics. We retrospectively assembled a dataset comprised of patients with paired ECG and HbA1c data. The population of patients who receive both an ECG and HbA1c may a biased sample of the complete outpatient population, so we adjusted the importance placed on each patient to generate a more representative pseudo-population. We found ECG-based assessment outperforms the ADA Risk test, achieving a higher area under the curve (0.80 vs. 0.68) and positive predictive value (13% vs. 9%) -- 2.6 times the prevalence of diabetes in the cohort. The AI-enhanced ECG significantly outperforms electrophysiologist interpretation of the ECG, suggesting that the task is beyond current clinical capabilities. Given the prevalence of ECGs in clinics and via wearable devices, such a tool would make precise, automated diabetes assessment widely accessible.