Edge-device co-inference, which concerns the cooperation between edge devices and an edge server for completing inference tasks over wireless networks, has been a promising technique for enabling various kinds of intelligent services at the network edge, e.g., auto-driving. In this paradigm, the concerned design objective of the network shifts from the traditional communication throughput to the effective and efficient execution of the inference task underpinned by the network, measured by, e.g., the inference accuracy and latency. In this paper, a task-oriented over-the-air computation scheme is proposed for a multidevice artificial intelligence system. Particularly, a novel tractable inference accuracy metric is proposed for classification tasks, which is called minimum pair-wise discriminant gain. Unlike prior work measuring the average of all class pairs in feature space, it measures the minimum distance of all class pairs. By maximizing the minimum pair-wise discriminant gain instead of its average counterpart, any pair of classes can be better separated in the feature space, and thus leading to a balanced and improved inference accuracy for all classes. Besides, this paper jointly optimizes the minimum discriminant gain of all feature elements instead of separately maximizing that of each element in the existing designs. As a result, the transmit power can be adaptively allocated to the feature elements according to their different contributions to the inference accuracy, opening an extra degree of freedom to improve inference performance. Extensive experiments are conducted using a concrete use case of human motion recognition to verify the superiority of the proposed design over the benchmarking scheme.

The performance of ML models degrades when the training population is different from that seen under operation. Towards assessing distributional robustness, we study the worst-case performance of a model over all subpopulations of a given size, defined with respect to core attributes Z. This notion of robustness can consider arbitrary (continuous) attributes Z, and automatically accounts for complex intersectionality in disadvantaged groups. We develop a scalable yet principled two-stage estimation procedure that can evaluate the robustness of state-of-the-art models. We prove that our procedure enjoys several finite-sample convergence guarantees, including dimension-free convergence. Instead of overly conservative notions based on Rademacher complexities, our evaluation error depends on the dimension of Z only through the out-of-sample error in estimating the performance conditional on Z. On real datasets, we demonstrate that our method certifies the robustness of a model and prevents deployment of unreliable models.

Emulating the mapping between quantities of interest and their control parameters using surrogate models finds widespread application in engineering design, including in numerical optimization and uncertainty quantification. Gaussian process models can serve as a probabilistic surrogate model of unknown functions, thereby making them highly suitable for engineering design and decision-making in the presence of uncertainty. In this work, we are interested in emulating quantities of interest observed from models of a system at multiple fidelities, which trade accuracy for computational efficiency. Using multifidelity Gaussian process models, to efficiently fuse models at multiple fidelities, we propose a novel method to actively learn the surrogate model via leave-one-out cross-validation (LOO-CV). Our proposed multifidelity cross-validation (\texttt{MFCV}) approach develops an adaptive approach to reduce the LOO-CV error at the target (highest) fidelity, by learning the correlations between the LOO-CV at all fidelities. \texttt{MFCV} develops a two-step lookahead policy to select optimal input-fidelity pairs, both in sequence and in batches, both for continuous and discrete fidelity spaces. We demonstrate the utility of our method on several synthetic test problems as well as on the thermal stress analysis of a gas turbine blade.

We present OxonFair, a new open source toolkit for enforcing fairness in binary classification. Compared to existing toolkits: (i) We support NLP and Computer Vision classification as well as standard tabular problems. (ii) We support enforcing fairness on validation data, making us robust to a wide range of overfitting challenges. (iii) Our approach can optimize any measure based on True Positives, False Positive, False Negatives, and True Negatives. This makes it easily extendable and much more expressive than existing toolkits. It supports 9/9 and 10/10 of the decision-based group metrics of two popular review papers. (iv) We jointly optimize a performance objective. This not only minimizes degradation while enforcing fairness, but can improve the performance of otherwise inadequately tuned unfair baselines. OxonFair is compatible with standard ML toolkits including sklearn, Autogluon, and PyTorch and is available online at https://github.com/oxfordinternetinstitute/oxonfair

Website Fingerprinting (WF) attacks identify the websites visited by users by performing traffic analysis, compromising user privacy. Particularly, DL-based WF attacks demonstrate impressive attack performance. However, the effectiveness of DL-based WF attacks relies on the collected complete and pure traffic during the page loading, which impacts the practicality of these attacks. The WF performance is rather low under dynamic network conditions and various WF defenses, particularly when the analyzed traffic is only a small part of the complete traffic. In this paper, we propose Holmes, a robust and reliable early-stage WF attack. Holmes utilizes temporal and spatial distribution analysis of website traffic to effectively identify websites in the early stages of page loading. Specifically, Holmes develops adaptive data augmentation based on the temporal distribution of website traffic and utilizes a supervised contrastive learning method to extract the correlations between the early-stage traffic and the pre-collected complete traffic. Holmes accurately identifies traffic in the early stages of page loading by computing the correlation of the traffic with the spatial distribution information, which ensures robust and reliable detection according to early-stage traffic. We extensively evaluate Holmes using six datasets. Compared to nine existing DL-based WF attacks, Holmes improves the F1-score of identifying early-stage traffic by an average of 169.18%. Furthermore, we replay the traffic of visiting real-world dark web websites. Holmes successfully identifies dark web websites when the ratio of page loading on average is only 21.71%, with an average precision improvement of 169.36% over the existing WF attacks.

Large language models (LLMs) (e.g., BERT [18], LLaMA [59]) which represents a direction of ever-increasing model sizes pre-trained on larger corpora, have demonstrated powerful capabilities in solving natural language processing (NLP) tasks, including question answering [99], text generation [100] and document understanding [101]. There are no clear and static thresholds regarding the model sizes. Early LLMs (e.g., BERT, RoBERTa) adopt an encoder architecture and show capabilities in text representation learning and natural language understanding. In recent years, more focus has been given to larger encoder-decoder [102] or decoder-only [103] architectures. As the model size scales up, such LLMs have also shown reasoning ability and even more advanced emergent ability [104], exposing a strong potential for Artificial General Intelligence (AGI). This inflection point, with the arrival of LLMs, marks a paradigm shift from explicit knowledge representation to a renewed focus on the hybrid representation of both explicit knowledge and parametric knowledge. As a popular approach for explicit knowledge representation, KGs are now widely investigated for the combination with Transformer-based LLMs, including pretrained masked language models (PLMs) like BERT and RoBERTa, and more recent generative LLMs like the GPT series and LLaMA. Some works use LLMs to augment knowledge graph representation learning.

Grounding DINO and the Segment Anything Model (SAM) have achieved impressive performance in zero-shot object detection and image segmentation, respectively. Together, they have a great potential to revolutionize applications in zero-shot semantic segmentation or data annotation. Yet, in specialized domains like medical image segmentation, objects of interest (e.g., organs, tissues, and tumors) may not fall in existing class names. To address this problem, the referring expression comprehension (REC) ability of Grounding DINO is leveraged to detect arbitrary targets by their language descriptions. However, recent studies have highlighted severe limitation of the REC framework in this application setting owing to its tendency to make false positive predictions when the target is absent in the given image. And, while this bottleneck is central to the prospect of open-set semantic segmentation, it is still largely unknown how much improvement can be achieved by studying the prediction errors. To this end, we perform empirical studies on six publicly available datasets across different domains and reveal that these errors consistently follow a predictable pattern and can, thus, be mitigated by a simple strategy. Specifically, we show that false positive detections with appreciable confidence scores generally occupy large image areas and can usually be filtered by their relative sizes. More importantly, we expect these observations to inspire future research in improving REC-based detection and automated segmentation. Meanwhile, we evaluate the performance of SAM on multiple datasets from various specialized domains and report significant improvements in segmentation performance and annotation time savings over manual approaches.

In this work, we improve the accuracy of several known algorithms to address the classification of large datasets when few labels are available. Our framework lies in the realm of graph-based semi-supervised learning. With novel modifications on Gaussian Random Fields Learning and Poisson Learning algorithms, we increase the accuracy and create more robust algorithms. Experimental results demonstrate the efficiency and superiority of the proposed methods over conventional graph-based semi-supervised techniques, especially in the context of imbalanced datasets.

Clinical trials are pivotal for developing new medical treatments, yet they typically pose some risks such as patient mortality, adverse events, and enrollment failure that waste immense efforts spanning over a decade. Applying artificial intelligence (AI) to forecast or simulate key events in clinical trials holds great potential for providing insights to guide trial designs. However, complex data collection and question definition requiring medical expertise and a deep understanding of trial designs have hindered the involvement of AI thus far. This paper tackles these challenges by presenting a comprehensive suite of meticulously curated AIready datasets covering multi-modal data (e.g., drug molecule, disease code, text, categorical/numerical features) and 8 crucial prediction challenges in clinical trial design, encompassing prediction of trial duration, patient dropout rate, serious adverse event, mortality rate, trial approval outcome, trial failure reason, drug dose finding, design of eligibility criteria. Furthermore, we provide basic validation methods for each task to ensure the datasets' usability and reliability. We anticipate that the availability of such open-access datasets will catalyze the development of advanced AI approaches for clinical trial design, ultimately advancing clinical trial research and accelerating medical solution development.

Mathematical problem solving is an important skill for Large Language Models (LLMs), both as an important capability and a proxy for a range of reasoning abilities. Existing benchmarks probe a diverse set of skills, but they yield aggregate accuracy metrics, obscuring specific abilities or weaknesses. Furthermore, they are difficult to extend with new problems, risking data contamination over time. To address these challenges, we propose MathCAMPS: a method to synthesize high-quality mathematical problems at scale, grounded on 44 fine-grained "standards" from the Mathematics Common Core (CC) Standard for K-8 grades. We encode each standard in a formal grammar, allowing us to sample diverse symbolic problems and their answers. We then use LLMs to realize the symbolic problems into word problems. We propose a cycle-consistency method for validating problem faithfulness. Finally, we derive follow-up questions from symbolic structures and convert them into follow-up word problems - a novel task of mathematical dialogue that probes for robustness in understanding. Experiments on 23 LLMs show surprising failures even in the strongest models (in particular when asked simple follow-up questions). Moreover, we evaluate training checkpoints of Pythia 12B on MathCAMPS, allowing us to analyze when particular mathematical skills develop during its training. Our framework enables the community to reproduce and extend our pipeline for a fraction of the typical cost of building new high-quality datasets.