arXiv subjects

Georgios Kaissis

Publications and source records attributed to Georgios Kaissis.

At least 19 recordsLinked to original sources

arXiv2026

Generalist-Specialist Mixture-of-Experts for Rare Pathology Detection in Multimodal Imaging

AI models for multimodal medical imaging must balance modality-specific specialization with cross-modal shared representations, a trade-off that pure Mixture-of-Experts (MoE) architectures currently fail to satisfy. Expert-based routing improves in-domain learning but may sacrifice cross-modal signals, which appear particularly important for rare (low-prevalence) pathologies in our experiments. To resolve this, we introduce Generalist-Specialist-MoE (GS-MoE), a two-branch (MoE) architecture that couples a cross-modal generalist model with distinct modality-specific specialists (experts) via domain-constrained feature fusion. On RadImageNet (1.35M images, 165 pathologies, three modalities), GS-MoE recovers detection of six low-prevalence pathologies on which every baseline scores F1 $=$ 0, with per-class gains up to +0.60 F1. It attains this while even slightly exceeding dense and specialist-only MoE aggregate baselines (MCC 0.770), while using ${\sim}53\%$ fewer active parameters at inference than the strongest investigated dense model.

cs.CV↗

arXiv2026

Memorisation bias in medical AI

Medical AI models hold immense potential to improve patient outcomes, but they are also known to unintentionally memorise individual records from their training datasets. While such memorisation has been linked to targeted privacy attacks, its consequences for clinical deployment, where patients may be assessed by a model that saw their historical data during training, remain poorly understood. Here we show that predictions on a patient's unseen future data can change significantly if a model observed that same patient's anonymised historical data during training, a phenomenon we term "memorisation bias". We demonstrate that this bias exists across diverse data modalities and model architectures, and over prolonged time spans: in some cases, memorisation bias persists on future records acquired decades after the historical records used for training. Moreover, in simulated prospective deployment, memorisation bias has asymmetric effects on the diagnostic accuracy of returning data contributors. When a patient returned with a de novo condition absent from their historical records in the training dataset, diagnostic sensitivity decreased significantly compared to an otherwise identical model not trained on their historical data. Conversely, when their health state was unchanged, both sensitivity and specificity were significantly inflated. Our findings reveal a previously uncharacterised risk in medical AI that arises when a model is deployed on patients who contributed to its training data. This exposes a shortcoming of current model development practice: the de-identification measures designed to protect patients' privacy make it difficult to identify returning contributors and exclude them from the AI-assisted interpretation of their own future data. Mitigating memorisation risks may thus require changes to current model training and deployment protocols.

cs.LG↗

arXiv2026

It depends: Incorporating correlations for joint aleatoric and epistemic uncertainties of high-dimensional output spaces

Uncertainty Quantification (UQ) plays a vital role in enhancing the reliability of deep learning model predictions, especially in scenarios with high-dimensional output spaces. This paper addresses the dual nature of uncertainty -- aleatoric and epistemic -- focusing on their joint integration in high-dimensional regression tasks. For example, in applications like medical image segmentation or restoration, aleatoric uncertainty captures inherent data noise, while epistemic uncertainty quantifies the model's confidence in unfamiliar conditions. Modeling both jointly enables more reliable predictions by reflecting both unavoidable variability and knowledge gaps, whereas modeling only one limits transparency and robustness. We propose a novel approach that approximates the resulting joint uncertainty using a low-rank plus diagonal covariance structure, capturing essential output correlations while avoiding the computational burdens of full covariance matrices. Unlike prior work, our method explicitly combines aleatoric and epistemic uncertainties into a unified second-order distribution that supports robust downstream analyses like sampling and log-likelihood evaluation. We further introduce stabilization strategies for efficient training and inference, achieving superior UQ in the tasks of image inpainting, colorization, optical flow, and depth estimation.

cs.LG↗

arXiv2026

Depth-adaptive Inference of Looped Language Models via Continuous Depth Batching

A main promise of looped language models (LMs) is depth-adaptive inference. By iterating a block of shared layers a variable number of times, the model can use less compute for "easy" tokens and more for "hard" ones. However, this adaptivity breaks standard batching: tokens in the same batch now require a different number of loops, so there is no unified forward pass, making efficient inference difficult. Standard inference frameworks like vLLM schedule on the token level and cannot handle this because tokens need to be removed from the batch within the forward pass. Loop-level scheduling has been proposed as a solution, but never implemented end to end. The key challenge is that looped architectures also contain non-looped boundary stages (e.g., token embedding and LM head) that must be scheduled at different frequencies than the loop. We introduce continuous depth batching (CDB), which schedules at the granularity of individual loop iterations. CDB handles boundary stages and loop steps in separate priority queues, makes exit decisions one step ahead, and overlaps all scheduling work with GPU computation. On Ouro 1.4B and Huginn 3.5B, CDB can realize up to $99\%$ of the theoretical maximum speed-up from adaptive-depth, translating to $1.5$-$1.9\times$ higher offline throughput and $45$-$90\%$ lower normalized latency under dynamic serving load.

cs.LG↗

arXiv2026

Addressing Benchmarking Gaps in Large Language Models for Health and Medicine with Dynamic Red-Teaming

Large language models (LLMs) are increasingly used to answer health-related questions and support healthcare workflows, yet evidence for their safety still relies heavily on static benchmarks that can rapidly become obsolete or be optimized against. Here we introduce a Dynamic, Automatic, and Systematic (DAS) red-teaming audit framework that continuously stress-tests LLMs for health across four safety-critical axes: robustness, privacy, bias/fairness, and hallucination/factual inaccuracies. Validated against board-certified clinicians with high concordance, a suite of adversarial agents autonomously mutates health-related test cases to uncover vulnerabilities in real time. Applying DAS to 15 proprietary and open-source LLMs revealed a profound gap between high static benchmark performance and low dynamic reliability--the "Benchmarking Gap". Despite median MedQA accuracy exceeding 80\%, 94\% of previously correct answers failed under dynamic robustness testing. This brittleness generalized to the realistic, open-ended HealthBench dataset, where top-tier models exhibited failure rates exceeding 70\% and sharp shifts in model rankings across evaluations, suggesting that high scores on established static benchmarks may reflect superficial memorization. We observed similarly high failure rates across other domains: privacy leaks were elicited in 86\% of scenarios, cognitive-bias priming altered recommendations in 81\% of fairness tests, and hallucination rates exceeded 74\% in widely used models. By converting LLM safety evaluation for health from a static checklist into a living adversarial audit, DAS provides a scalable framework for surfacing latent risks before such systems are deployed in consumer-facing health assistants, clinician-facing tools, and broader healthcare workflows. Code is available at https://github.com/JZPeterPan/DAS-Medical-Red-Teaming-Agents.

cs.LG↗

arXiv2026

Gaussian DP for Reporting Differential Privacy Guarantees in Machine Learning

Current practices for reporting differential privacy (DP) guarantees for machine learning (ML) algorithms such as DP-SGD provide an incomplete and potentially misleading picture. For instance, if only a single $(\varepsilon, δ)$ is known about a mechanism, standard analyses show that there could exist highly accurate inference attacks against training data records, when, upon a more careful analysis, such accurate attacks do not exist for most practical mechanisms. In this position paper, we argue that using _non-asymptotic_ Gaussian Differential Privacy (GDP) as the primary means of communicating DP guarantees in ML avoids these potential downsides. Using two recent developments in the DP literature: (i) open-source numerical accountants capable of computing the privacy profile and $f$-DP curves of DP-SGD to arbitrary accuracy, and (ii) a decision-theoretic metric over DP representations, we show how to provide non-asymptotic bounds on GDP using numerical accountants, and show that GDP can capture the entire privacy profile of DP-SGD and related algorithms with virtually no error, as quantified by the metric. To support our claims, we investigate the privacy profiles of state-of-the-art DP large-scale image classification, and the TopDown algorithm for the U.S. Decennial Census, observing that GDP fits their profiles remarkably well in all cases. We conclude with a discussion on the strengths and weaknesses of this approach, and discuss which other privacy mechanisms could benefit from GDP.

cs.LG↗

arXiv2026

Optimal conversion from Rényi Differential Privacy to $f$-Differential Privacy

We prove the conjecture stated in Appendix F.3 of \citet{zhu2022optimalaccountingdifferentialprivacy}: among all conversion rules that map a Rényi Differential Privacy (RDP) profile $τ\mapsto ρ(τ)$ to a valid hypothesis-testing trade-off $f$, the rule based on the intersection of single-order RDP privacy regions is optimal. This optimality holds simultaneously for all valid RDP profiles and for all Type I error levels $α$. Concretely, we show that in the space of trade-off functions, the tightest possible bound is $f_{ρ(\cdot)}(α) = \sup_{τ\geq 0.5} f_{τ,ρ(τ)}(α)$: the pointwise maximum of the single-order bounds for each RDP privacy region. Our proof unifies and sharpens the insights of \citet{balle2019hypothesistestinginterpretationsrenyi}, \citet{asoodeh2021variantsdifferentialprivacylossless}, and \citet{zhu2022optimalaccountingdifferentialprivacy}. Our analysis relies on a precise geometric characterization of the RDP privacy region, leveraging its convexity and the fact that its boundary is determined exclusively by Bernoulli mechanisms. Our results establish that the \enquote{intersection-of-RDP-privacy-regions} rule is not only valid, but optimal: no other black-box conversion can uniformly dominate it in the Blackwell sense, marking the fundamental limit of what can be inferred about a mechanism's privacy solely from its RDP guarantees.

cs.CR↗

arXiv2026

Efficient numeracy in language models through single-token number embeddings

To drive progress in science and engineering, large language models (LLMs) must be able to process large amounts of numerical data and solve long calculations efficiently. This is currently only possible through the use of external tools or extensive reasoning chains, either weakening the numerical representations of LLMs or limiting the length of problems they can solve. We show that frontier LLMs require excessive amounts of reasoning tokens to solve even basic calculations, which is exacerbated by their tokenization strategies that split single numbers into multiple tokens. This motivates the need for efficient and effective single-token number encodings. We introduce a set of desiderata for such encodings and show that existing approaches fail to fulfill them. To address these shortcomings, we propose BitTokens, a novel encoding strategy that represents any number as a single token using its IEEE 754 binary floating-point representation. Through extensive experiments we show that our BitTokens allow even small language models to learn algorithms that solve basic arithmetic operations nearly perfectly. This newly gained efficiency could expand the length and complexity of problems language models can solve.

cs.LG↗

arXiv2026

Survival In-Context: Amortized Bayesian Survival Analysis via Prior-Fitted Networks

Survival analysis is crucial for many medical applications, but remains challenging for modern machine learning due to limited data, censoring, and the heterogeneity of tabular covariates. While the prior-fitted paradigm, which relies on pretraining models on large collections of synthetic datasets, has recently facilitated tabular foundation models for classification and regression, its suitability for time-to-event modeling remains unclear. We propose a flexible survival data generation framework that defines a rich survival prior with explicit control over covariates and time-event distributions. Building on this prior, we introduce Survival In-Context (SIC), a prior-fitted in-context learning model for survival analysis that is pretrained exclusively on synthetic data. SIC is trained to approximate Bayesian posterior predictive inference under the synthetic survival prior, enabling individualized survival prediction in a single forward pass, requiring no task-specific training or hyperparameter tuning. Across a broad evaluation on real-world survival datasets, SIC achieves competitive or superior performance compared to classical and deep survival models, particularly in small and medium-sized data regimes, highlighting the promise of a prior-fitted paradigm for survival analysis. The code and pretrained models will be made available upon publication.

cs.LG↗

arXiv2026

How Much Is One Recurrence Worth? Iso-Depth Scaling Laws for Looped Language Models

We measure how much one recurrence is worth to a looped (depth-recurrent) transformer, in equivalent unique parameters. From an iso-depth pretraining sweep across recurrence counts $r \in \{1, 2, 4, 8\}$ spanning ${\sim}50\times$ in training compute, we fit a joint scaling law $L = E + A\,(N_\text{once} + r^φ N_\text{rec})^{-α} + B\,D^{-β}$ and measure a recurrence-equivalence exponent $φ= 0.46$. Intuitively, $φ$ tells us whether looping a block $r$ times is equivalent in validation loss to $r$ unique blocks of a non-looped model (full equivalence, $φ{=}1$) or to a single block run repeatedly with no capacity gain ($φ{=}0$). Our $φ= 0.46$ sits in between, so replacing unique blocks with shared recurrences increases validation loss at matched training compute. For example, at $r{=}4$ a 410M looped model performs on par with a 580M non-looped model, but incurs the training cost of a 1B non-looped one. We demonstrate the utility of $φ$ as a diagnostic tool on two case studies: commonly used truncated backpropagation lowers $φ$ to $0.38$, indicating that the loop mechanism is poorly trained under truncation, even though validation loss decreases. Conversely, hyperconnections raise $φ$ to $0.65$, a genuine capacity gain. Our method separates true loop improvements from training-side gains, a distinction raw validation loss cannot make.

cs.LG↗

arXiv2026

Is your algorithm unlearning or untraining?

As models are getting larger and are trained on increasing amounts of data, there has been an explosion of interest into how we can ``delete'' specific data points or behaviours from a trained model, after the fact. This goal has been referred to as ``machine unlearning''. In this note, we argue that the term ``unlearning'' has been overloaded, with different research efforts spanning two distinct problem formulations, but without that distinction having been observed or acknowledged in the literature. This causes various issues, including ambiguity around when an algorithm is expected to work, use of inappropriate metrics and baselines when comparing different algorithms to one another, difficulty in interpreting results, as well as missed opportunities for pursuing critical research directions. In this note, we address this issue by establishing a fundamental distinction between two notions that we identify as \unlearning and \untraining, illustrated in Figure 1. In short, \untraining aims to reverse the effect of having trained on a given forget set, i.e. to remove the influence that that specific forget set examples had on the model during training. On the other hand, the goal of \unlearning is not just to remove the influence of those given examples, but to use those examples for the purpose of more broadly removing the entire underlying distribution from which those examples were sampled (e.g. the concept or behaviour that those examples represent). We discuss technical definitions of these problems and map problem settings studied in the literature to each. We hope to initiate discussions on disambiguating technical definitions and identify a set of overlooked research questions, as we believe that this a key missing step for accelerating progress in the field of ``unlearning''.

cs.LG↗

arXiv2026

On Arbitrary Predictions from Equally Valid Models

Model multiplicity refers to the existence of multiple machine learning models that describe the data equally well but may produce different predictions on individual samples. In medicine, these models can admit conflicting predictions for the same patient -- a risk that is poorly understood and insufficiently addressed. In this study, we empirically analyze the extent, drivers, and ramifications of predictive multiplicity across diverse medical tasks and model architectures, and show that even small ensembles can mitigate/eliminate predictive multiplicity in practice. Our analysis reveals that (1) standard validation metrics fail to identify a uniquely optimal model and (2) a substantial amount of predictions hinges on arbitrary choices made during model development. Using multiple models instead of a single model reveals instances where predictions differ across equally plausible models -- highlighting patients that would receive arbitrary diagnoses if any single model were used. In contrast, (3) a small ensemble paired with an abstention strategy can effectively mitigate measurable predictive multiplicity in practice; predictions with high inter-model consensus may thus be amenable to automated classification. While accuracy is not a principled antidote to predictive multiplicity, we find that (4) higher accuracy achieved through increased model capacity reduces predictive multiplicity. Our findings underscore the clinical importance of accounting for model multiplicity and advocate for ensemble-based strategies to improve diagnostic reliability. In cases where models fail to reach sufficient consensus, we recommend deferring decisions to expert review.

cs.LG↗

arXiv2026

Step-resolved data attribution for looped transformers

We study how individual training examples shape the internal computation of looped transformers, where a shared block is applied for $τ$ recurrent iterations to enable latent reasoning. Existing training-data influence estimators such as TracIn yield a single scalar score that aggregates over all loop iterations, obscuring when during the recurrent computation a training example matters. We introduce \textit{Step-Decomposed Influence (SDI)}, which decomposes TracIn into a length-$τ$ influence trajectory by unrolling the recurrent computation graph and attributing influence to specific loop iterations. To make SDI practical at transformer scale, we propose a TensorSketch implementation that never materialises per-example gradients. Experiments on looped GPT-style models and algorithmic reasoning tasks show that SDI scales excellently, matches full-gradient baselines with low error and supports a broad range of data attribution and interpretability tasks with per-step insights into the latent reasoning process.

cs.LG↗

arXiv2026

Unifying Re-Identification, Attribute Inference, and Data Reconstruction Risks in Differential Privacy

Differentially private (DP) mechanisms are difficult to interpret and calibrate because existing methods for mapping standard privacy parameters to concrete privacy risks -- re-identification, attribute inference, and data reconstruction -- are both overly pessimistic and inconsistent. In this work, we use the hypothesis-testing interpretation of DP ($f$-DP), and determine that bounds on attack success can take the same unified form across re-identification, attribute inference, and data reconstruction risks. Our unified bounds are (1) consistent across a multitude of attack settings, and (2) tunable, enabling practitioners to evaluate risk with respect to arbitrary, including worst-case, levels of baseline risk. Empirically, our results are tighter than prior methods using $\varepsilon$-DP, Rényi DP, and concentrated DP. As a result, calibrating noise using our bounds can reduce the required noise by 20% at the same risk level, which yields, e.g., an accuracy increase from 52% to 70% in a text classification task. Overall, this unifying perspective provides a principled framework for interpreting and calibrating the degree of protection in DP against specific levels of re-identification, attribute inference, or data reconstruction risk.

cs.LG↗

arXiv2026

Redirection for Erasing Memory (REM): Towards a universal unlearning method for corrupted data

Machine unlearning is studied for a multitude of tasks, but specialization of unlearning methods to particular tasks has made their systematic comparison challenging. To address this issue, we propose a conceptual space to characterize diverse corrupted data unlearning tasks in vision classifiers. This space is described by two dimensions, the discovery rate (the fraction of the corrupted data that are known at unlearning time) and the statistical regularity of the corrupted data (from random exemplars to shared concepts). Methods proposed previously have been targeted at portions of this space and-we show-fail predictably outside these regions. We propose a novel method, Redirection for Erasing Memory (REM), whose key feature is that corrupted data are redirected to dedicated neurons introduced at unlearning time and then discarded or deactivated to suppress the influence of corrupted data. REM performs strongly across the space of tasks, in contrast to prior SOTA methods that fail outside the regions for which they were designed.

cs.LG↗

arXiv2026

Unintended Memorization of Sensitive Information in Fine-Tuned Language Models

Fine-tuning Large Language Models (LLMs) on sensitive datasets carries a substantial risk of unintended memorization and leakage of Personally Identifiable Information (PII), which can violate privacy regulations and compromise individual safety. In this work, we systematically investigate a critical and underexplored vulnerability: the exposure of PII that appears only in model inputs, not in training targets. Using both synthetic and real-world datasets, we design controlled extraction probes to quantify unintended PII memorization and study how factors such as language, PII frequency, task type, and model size influence memorization behavior. We further benchmark four privacy-preserving approaches including differential privacy, machine unlearning, regularization, and preference alignment, evaluating their trade-offs between privacy and task performance. Our results show that post-training methods generally provide more consistent privacy-utility trade-offs, while differential privacy achieves strong reduction in leakage in specific settings, although it can introduce training instability. These findings highlight the persistent challenge of memorization in fine-tuned LLMs and emphasize the need for robust, scalable privacy-preserving techniques.

cs.LG↗

arXiv2026

Your Privacy Depends on Others: Collusion Vulnerabilities in Individual Differential Privacy

Individual Differential Privacy (iDP) promises users control over their privacy, but this promise can be broken in practice. We reveal a previously overlooked vulnerability in sampling-based iDP mechanisms: while conforming to the iDP guarantees, an individual's privacy risk is not solely governed by their own privacy budget, but critically depends on the privacy choices of all other data contributors. This creates a mismatch between the promise of individual privacy control and the reality of a system where risk is collectively determined. We demonstrate empirically that certain distributions of privacy preferences can unintentionally inflate the privacy risk of individuals, even when their formal guarantees are met. Moreover, this excess risk provides an exploitable attack vector. A central adversary or a set of colluding adversaries can deliberately choose privacy budgets to amplify vulnerabilities of targeted individuals. Most importantly, this attack operates entirely within the guarantees of DP, hiding this excess vulnerability. Our empirical evaluation demonstrates successful attacks against 62% of targeted individuals, substantially increasing their membership inference susceptibility. To mitigate this, we propose $(\varepsilon_i,δ_i,\overlineΔ)$-iDP a privacy contract that uses $Δ$-divergences to provide users with a hard upper bound on their excess vulnerability, while offering flexibility to mechanism design. Our findings expose a fundamental challenge to the current paradigm, demanding a re-evaluation of how iDP systems are designed, audited, communicated, and deployed to make excess risks transparent and controllable.

cs.CR↗

arXiv2025

The Hitchhiker's Guide to Efficient, End-to-End, and Tight DP Auditing

In this paper, we systematize research on auditing Differential Privacy (DP) techniques, aiming to identify key insights and open challenges. First, we introduce a comprehensive framework for reviewing work in the field and establish three cross-contextual desiderata that DP audits should target -- namely, efficiency, end-to-end-ness, and tightness. Then, we systematize the modes of operation of state-of-the-art DP auditing techniques, including threat models, attacks, and evaluation functions. This allows us to highlight key details overlooked by prior work, analyze the limiting factors to achieving the three desiderata, and identify open research problems. Overall, our work provides a reusable and systematic methodology geared to assess progress in the field and identify friction points and future directions for our community to focus on.

cs.CR↗