The fourth Machine Learning for Health (ML4H) symposium was held in person on December 15th and 16th, 2024, in the traditional, ancestral, and unceded territories of the Musqueam, Squamish, and Tsleil-Waututh Nations in Vancouver, British Columbia, Canada. The symposium included research roundtable sessions to foster discussions between participants and senior researchers on timely and relevant topics for the ML4H community. The organization of the research roundtables at the conference involved 13 senior and 27 junior chairs across 13 tables. Each roundtable session included an invited senior chair (with substantial experience in the field), junior chairs (responsible for facilitating the discussion), and attendees from diverse backgrounds with an interest in the session's topic.

Dense object detection is widely used in automatic driving, video surveillance, and other fields. This paper focuses on the challenging task of dense object detection. Currently, detection methods based on greedy algorithms, such as non-maximum suppression (NMS), often produce many repetitive predictions or missed detections in dense scenarios, which is a common problem faced by NMS-based algorithms. Through the end-to-end DETR (DEtection TRansformer), as a type of detector that can incorporate the post-processing de-duplication capability of NMS, etc., into the network, we found that homogeneous queries in the query-based detector lead to a reduction in the de-duplication capability of the network and the learning efficiency of the encoder, resulting in duplicate prediction and missed detection problems. To solve this problem, we propose learnable differentiated encoding to de-homogenize the queries, and at the same time, queries can communicate with each other via differentiated encoding information, replacing the previous self-attention among the queries. In addition, we used joint loss on the output of the encoder that considered both location and confidence prediction to give a higher-quality initialization for queries. Without cumbersome decoder stacking and guaranteeing accuracy, our proposed end-to-end detection framework was more concise and reduced the number of parameters by about 8% compared to deformable DETR. Our method achieved excellent results on the challenging CrowdHuman dataset with 93.6% average precision (AP), 39.2% MR-2, and 84.3% JI. The performance overperformed previous SOTA methods, such as Iter-E2EDet (Progressive End-to-End Object Detection) and MIP (One proposal, Multiple predictions). In addition, our method is more robust in various scenarios with different densities.

The application of AI in healthcare allows for the identification of complex patterns in patient data, improving diagnostic accuracy, treatment personalization, and operational efficiency [1]. Healthcare providers are increasingly leveraging predictive analytics to foresee health outcomes, enabling earlier interventions and more targeted care [2][26]. For instance, AI models have proven effective in identifying high-risk patients and optimizing preventive care strategies [3]. Diabetes, a major global health challenge, requires early detection and preventive care. Predictive models built using accessible tools like BigQuery ML can help healthcare professionals identify at-risk individuals efficiently. Cloud computing serves as a critical tool for AI and ML in healthcare, addressing many of the technical and infrastructural challenges associated with large-scale data analysis. With scalable infrastructure, cloud platforms allow healthcare providers to process and store vast amounts of data, facilitating AI-driven insights without the need of extensive on-site resources [4].

The implementation of Hyperdimensional Computing (HDC) on In-Memory Computing (IMC) architectures faces significant challenges due to the mismatch between highdimensional vectors and IMC array sizes, leading to inefficient memory utilization and increased computation cycles. This paper presents MEMHD, a Memory-Efficient Multi-centroid HDC framework designed to address these challenges. MEMHD introduces a clustering-based initialization method and quantization aware iterative learning for multi-centroid associative memory. Through these approaches and its overall architecture, MEMHD achieves a significant reduction in memory requirements while maintaining or improving classification accuracy. Our approach achieves full utilization of IMC arrays and enables one-shot (or few-shot) associative search. Experimental results demonstrate that MEMHD outperforms state-of-the-art binary HDC models, achieving up to 13.69% higher accuracy with the same memory usage, or 13.25x more memory efficiency at the same accuracy level. Moreover, MEMHD reduces computation cycles by up to 80x and array usage by up to 71x compared to baseline IMC mapping methods when mapped to 128x128 IMC arrays, while significantly improving energy and computation cycle efficiency.

Semivalue-based data valuation in machine learning (ML) quantifies the contribution of individual data points to a downstream ML task by leveraging principles from cooperative game theory and the notion of utility. While this framework has been used in practice for assessing data quality, our experiments reveal inconsistent valuation outcomes across different utilities, albeit all related to ML performance. Beyond raising concerns about the reliability of data valuation, this inconsistency is challenging to interpret, as it stems from the complex interaction of the utility with data points and semivalue weights, which has barely been studied in prior work. In this paper, we take a first step toward clarifying the utility impact on semivalue-based data valuation. Specifically, we provide geometric interpretations of this impact for a broad family of classification utilities, which includes the accuracy and the arithmetic mean. We introduce the notion of spatial signatures: given a semivalue, data points can be embedded into a two-dimensional space, and utility functions map to the dual of this space. This geometric perspective separates the influence of the dataset and semivalue from that of the utility, providing a theoretical explanation for the experimentally observed sensitivity of valuation outcomes to the utility choice.

High-risk artificial intelligence and machine learning classification tasks, such as healthcare diagnosis, require accurate and interpretable prediction models. However, classifier algorithms typically sacrifice individual case-accuracy for overall model accuracy, limiting analysis of class overlap areas regardless of task significance. The Adaptive Boosting meta-algorithm, which won the 2003 G\"odel Prize, analytically assigns higher weights to misclassified cases to reclassify. However, it relies on weaker base classifiers that are iteratively strengthened, limiting improvements from base classifiers. Combining visual and computational approaches enables selecting stronger base classifiers before boosting. This paper proposes moving boosting methodology from focusing on only misclassified cases to all cases in the class overlap areas using Computational and Interactive Visual Learning (CIVL) with a Human-in-the-Loop. It builds classifiers in lossless visualizations integrating human domain expertise and visual insights. A Divide and Classify process splits cases to simple and complex, classifying these individually through computational analysis and data visualization with lossless visualization spaces of Parallel Coordinates or other General Line Coordinates. After finding pure and overlap class areas simple cases in pure areas are classified, generating interpretable sub-models like decision rules in Propositional and First-order Logics. Only multidimensional cases in the overlap areas are losslessly visualized simplifying end-user cognitive tasks to identify difficult case patterns, including engineering features to form new classifiable patterns. Demonstration shows a perfectly accurate and losslessly interpretable model of the Iris dataset, and simulated data shows generalized benefits to accuracy and interpretability of models, increasing end-user confidence in discovered models.

The proliferation of IoT devices has significantly increased network vulnerabilities, creating an urgent need for effective Intrusion Detection Systems (IDS). Machine Learning-based IDS (ML-IDS) offer advanced detection capabilities but rely on labeled attack data, which limits their ability to identify unknown threats. Self-Supervised Learning (SSL) presents a promising solution by using only normal data to detect patterns and anomalies. This paper introduces SAFE, a novel framework that transforms tabular network intrusion data into an image-like format, enabling Masked Autoencoders (MAEs) to learn robust representations of network behavior. The features extracted by the MAEs are then incorporated into a lightweight novelty detector, enhancing the effectiveness of anomaly detection. Experimental results demonstrate that SAFE outperforms the state-of-the-art anomaly detection method, Scale Learning-based Deep Anomaly Detection method (SLAD), by up to 26.2% and surpasses the state-of-the-art SSL-based network intrusion detection approach, Anomal-E, by up to 23.5% in F1-score.

Membership Inference Attacks (MIAs) have emerged as a valuable framework for evaluating privacy leakage by machine learning models. Score-based MIAs are distinguished, in particular, by their ability to exploit the confidence scores that the model generates for particular inputs. Existing score-based MIAs implicitly assume that the adversary has access to the target model's hyperparameters, which can be used to train the shadow models for the attack. In this work, we demonstrate that the knowledge of target hyperparameters is not a prerequisite for MIA in the transfer learning setting. Based on this, we propose a novel approach to select the hyperparameters for training the shadow models for MIA when the attacker has no prior knowledge about them by matching the output distributions of target and shadow models. We demonstrate that using the new approach yields hyperparameters that lead to an attack near indistinguishable in performance from an attack that uses target hyperparameters to train the shadow models. Furthermore, we study the empirical privacy risk of unaccounted use of training data for hyperparameter optimization (HPO) in differentially private (DP) transfer learning. We find no statistically significant evidence that performing HPO using training data would increase vulnerability to MIA.

Model distillation -- using outputs from a large teacher model to teach a small student model -- is a practical means of creating efficient models for a particular task. We ask: Can we identify a students' teacher based on its outputs? Such "footprints" left by teacher LLMs would be interesting artifacts. Beyond this, reliable teacher inference may have practical implications as actors seek to distill specific capabilities of massive proprietary LLMs into deployed smaller LMs, potentially violating terms of service. We consider practical task distillation targets including summarization, question answering, and instruction-following. We assume a finite set of candidate teacher models, which we treat as blackboxes. We design discriminative models that operate over lexical features. We find that $n$-gram similarity alone is unreliable for identifying teachers, but part-of-speech (PoS) templates preferred by student models mimic those of their teachers.

Ischaemic stroke, a leading cause of death and disability, critically relies on neuroimaging for characterising the anatomical pattern of injury. Diffusion-weighted imaging (DWI) provides the highest expressivity in ischemic stroke but poses substantial challenges for automated lesion segmentation: susceptibility artefacts, morphological heterogeneity, age-related comorbidities, time-dependent signal dynamics, instrumental variability, and limited labelled data. Current U-Net-based models therefore underperform, a problem accentuated by inadequate evaluation metrics that focus on mean performance, neglecting anatomical, subpopulation, and acquisition-dependent variability. Here, we present a high-performance DWI lesion segmentation tool addressing these challenges through optimized vision transformer-based architectures, integration of 3563 annotated lesions from multi-site data, and algorithmic enhancements, achieving state-of-the-art results. We further propose a novel evaluative framework assessing model fidelity, equity (across demographics and lesion subtypes), anatomical precision, and robustness to instrumental variability, promoting clinical and research utility. This work advances stroke imaging by reconciling model expressivity with domain-specific challenges and redefining performance benchmarks to prioritize equity and generalizability, critical for personalized medicine and mechanistic research.