Background: Continuous glucose monitoring (CGM) has the potential to improve glucose control in critically ill patients, provided that its trend accuracy is reliable. We evaluated the trend accuracy of a subcutaneous CGM system (Dexcom G6) compared with intermittent arterial blood gas (ABG) measurements in intensive care unit (ICU) patients receiving insulin. Methods: We enrolled 40 adult ICU patients receiving insulin and organ-supportive therapies. We assessed trend accuracy using the Rate Error Grid Analysis (R-EGA) and the Diabetes Technology Society Trend Accuracy Matrix (DTS-TAM), overall, across different ABG levels, and over time from CGM initiation. Results: A total of 2701 paired CGM-ABG trends were analyzed, with a median (IQR) time difference between readings of 83 (65-125) minutes. Overall, 99.7% of trends were classified in R-EGA Zone A and 0.3% in Zone B. On DTS-TAM analysis, 98.6% of trends fell in the No Risk category, while 1.7% were in the adjacent Mild-to-Moderate Risk categories. Trends were more frequently categorized as Mild-to-Moderate Risk when ABG values were <100 mg/dL (5.56 mmol/L) (3.6%) compared with 100 to 180 mg/dL (5.56 to 10 mmol/L) (1.3%) or >180 mg/dL (10 mmol/L) (1.6%). During the first 24 hours of CGM use, 2.9% of trends fell into the Mild-to-Moderate Risk categories, compared with 0.9% beyond 24 hours. Conclusions: In critically ill patients receiving insulin, CGM demonstrated high overall trend accuracy relative to ABG. Trend accuracy was reduced at lower glucose ranges and during the initial 24 hours of CGM use.

Background: In the Intensive Care Unit (ICU), data stored in patient data management systems (PDMS) is commonly used in clinical practice and research. Parameters from point-of-care arterial blood gas (BG) analysis are used in the diagnosis and definition of syndromes such as sepsis and ARDS, but manual entry of the blood source (arterial or venous) into the PDMS introduces the risk of mislabeling venous samples as arterial. Our study aimed to employ supervised machine learning to accurately identify blood gas samples as arterial or venous using PDMS data.

Methods: A retrospective, single-center observational cohort study including all blood gases during 2018 from a Swedish, pediatric and adult general ICU. Chemical parameters from BG analysis and clinical parameters such as mean arterial pressure (MAP) and saturation (SpO2) were utilized as features. A specialist physician in Intensive Care manually determined the true class of each sample through comprehensive retrospective chart review. The samples were split into training, testing and holdout sets. Training was performed using cross-validation in the training set, with forward stepwise feature selection and Bayesian hyperparameter optimization, and accuracy was assessed using area under the precision recall curve (AUCPR) in the test set. The best model was compared to a multivariate logistic regression model (LR) in the holdout set.

Results: Among 33,800 samples (30,753 arterial, 3,047 non-arterial) from 691 ICU admissions, 150 (0.44%) were erroneously marked. The best performing algorithm was extreme gradient boosting (XGboost) using 9 features, with an AUCPR of 0.9974 (95% CI 0.9961–0.9984), significantly better than the LR model (AUCPR = 0.9791, 95% CI 0.9651–0.9904).

Conclusion: Supervised machine learning demonstrates efficacy in determining blood gas sample type from ICU patients. This approach shows promise for improving the accuracy of research and clinical applications relying on blood gas data.

In this paper, we study an information-theoretic problem of designing a fair representation that attains bounded statistical (demographic) parity. More specifically, an agent uses some useful data X to solve a task T. Since both X and T are correlated with some sensitive attribute or secret S, the agent designs a representation Y that satisfies a bounded statistical parity and/or privacy leakage constraint, that is, such that I(Y ; S) ≤ ϵ. Here, we relax the perfect demographic (statistical) parity and consider a bounded-parity constraint. In this work, we design the representation Y that maximizes the mutual information I(Y ; T) about the task while satisfying a bounded compression (or encoding rate) constraint, that is, ensuring that I(Y ; X) ≤ r. Simultaneously, Y satisfies the bounded statistical parity constraint I(Y ; S) ≤ ϵ. To design Y , we use extended versions of the Functional Representation Lemma and the Strong Functional Representation Lemma which are based on randomization techniques and study the tightness of the obtained bounds in special cases. The main idea to derive the lower bounds is to use randomization over useful data X or sensitive data S. Considering perfect demographic parity, i.e., ϵ = 0, we improve the existing results (lower bounds) by using a tighter version of the Strong Functional Representation Lemma and propose new upper bounds. We then propose upper and lower bounds for the main problem and show that allowing non-zero leakage can improve the attained utility. Finally, we study the bounds and compare them in a numerical example. The problem studied in this paper can also be interpreted as one of code design with bounded leakage and bounded rate privacy considering the sensitive attribute as a secret.

Hyperspherical Prototypical Learning (HPL) is a supervised approach to representation learning that designs class prototypes on the unit hypersphere. The prototypes bias the representations to class separation in a scale invariant and known geometry. Previous approaches to HPL have either of the following shortcomings: (i) they follow an unprincipled optimisation procedure; or (ii) they are theoretically sound, but are constrained to only one possible latent dimension. In this paper, we address both shortcomings. To address (i), we present a principled optimisation procedure whose solution we show is optimal. To address (ii), we construct well-separated prototypes in a wide range of dimensions using linear block codes. Additionally, we give a full characterisation of the optimal prototype placement in terms of achievable and converse bounds, showing that our proposed methods are near-optimal.

Hyperspherical Prototypical Learning (HPL) is a supervised approach to representation learning that designs class prototypes on the unit hypersphere. The prototypes bias the representations to class separation in a scale invariant and known geometry. Previous approaches to HPL have either of the following shortcomings: (i) they follow an unprincipled optimisation procedure; or (ii) they are theoretically sound, but are constrained to only one possible latent dimension. In this paper, we address both shortcomings. To address (i), we present a principled optimisation procedure whose solution we show is optimal. To address (ii), we construct well-separated prototypes in a wide range of dimensions using linear block codes. Additionally, we give a full characterisation of the optimal prototype placement in terms of achievable and converse bounds, showing that our proposed methods are near-optimal.

Hyperspherical Prototypical Learning (HPL) is a supervised approach to representation learning that designs class prototypes on the unit hypersphere. The prototypes bias the representations to class separation in a scale invariant and known geometry. Previous approaches to HPL have either of the following shortcomings: (i) they follow an unprincipled optimisation procedure; or (ii) they are theoretically sound, but are constrained to only one possible latent dimension. In this paper, we address both shortcomings. To address (i), we present a principled optimisation procedure whose solution we show is optimal. To address (ii), we construct well-separated prototypes in a wide range of dimensions using linear block codes. Additionally, we give a full characterisation of the optimal prototype placement in terms of achievable and converse bounds, showing that our proposed methods are near-optimal. GitHub: martinlindstrom/coding theoretic hpl

In this paper we propose an unpredictability-based jamming defense framework based on the principles of Moving Target Defense for a wireless communication problem. Taking advantage of the complex nature of large-scale cyber-physical systems, we consider a platform consisting of a single receiving component but multiple potential transmitting components, each equipped with a multi-antenna phased array. We formulate an optimization problem over the probability simplex that characterizes a randomized receiving angle which seeks to balance between the estimated performance of the transmission and an entropy-based unpredictability measure. Furthermore, we explore the effect of an intelligent adversary that has knowledge of the derived probabilities and optimally places a single-antenna jamming device to disrupt the communication links. Finally, simulation results showcase the efficacy of the proposed algorithm.

This paper studies the truncation method from Alquier [1] to derive high-probability PAC-Bayes bounds for unbounded losses with heavy tails. Assuming that the p-th moment is bounded, the resulting bounds interpolate between a slow rate 1/√n when p=2, and a fast rate 1/n when p→∞ and the loss is essentially bounded. Moreover, the paper derives a high-probability PAC-Bayes bound for losses with a bounded variance. This bound has an exponentially better dependence on the confidence parameter and the dependency measure than previous bounds in the literature. Finally, the paper extends all results to guarantees in expectation and single-draw PAC-Bayes. In order to so, it obtains analogues of the PAC-Bayes fast rate bound for bounded losses from [2] in these settings. The full version of the paper can be found in https://arxiv.org/abs/2403.16681.

In this paper, we present new high-probability PAC-Bayes bounds for different types of losses. Firstly, for losses with a bounded range, we recover a strengthened version of Catoni's bound that holds uniformly for all parameter values. This leads to new fast-rate and mixed-rate bounds that are interpretable and tighter than previous bounds in the literature. In particular, the fast-rate bound is equivalent to the Seeger-Langford bound. Secondly, for losses with more general tail behaviors, we introduce two new parameter-free bounds: a PAC-Bayes Chernoff analogue when the loss' cumulative generating function is bounded, and a bound when the loss' second moment is bounded. These two bounds are obtained using a new technique based on a discretization of the space of possible events for the "in probability" parameter optimization problem. This technique is both simpler and more general than previous approaches optimizing over a grid on the parameters' space. Finally, using a simple technique that is applicable to any existing bound, we extend all previous results to anytime-valid bounds.

We introduce a new class of codebook-aware jamming strategies against coded transmissions over AWGN channels. The proposed strategies derive attack vectors from non-zero positions of minimum-weight codewords in Hamming space and utilize the geometry of minimum-distance error events in Euclidean space in their attack. We characterize the success probability of the attacker analytically and utilize these results for attack optimization. We demonstrate that the proposed jamming attacks are highly efficient; compared to Gaussian attack vectors with the same energy budget, the attacker's success probability is increased by more than two orders of magnitude while only consuming a fraction of the energy of the attacked codeword.