Searcharxiv⌕ Search

arXiv subjects

Francesca Dominici

Publications and source records attributed to Francesca Dominici.

At least 19 recordsLinked to original sources

Re-Reading Is Not Free: Token Traces and Cache-Aware Energy Estimation for LLM Agents in Software Engineering

As large language models (LLMs) evolve from conversational assistants into autonomous agents embedded in software development workflows, estimating their environmental footprint requires moving beyond conventional token-based accounting. Existing energy estimators focus on isolated request-response interactions, overlooking the repeated processing of accumulated context, cache reuse, and work delegated to subagents. We analyze nearly one billion tokens collected across 42 sessions and 752 task episodes over ten weeks of real-world AI-assisted software development. Cached-context reads make up 94.8% of all API-reported tokens, while uncached input accounts for only 0.017%; for every generated token, approximately 260 cached tokens are read. Context grows 5.3-fold over a session compared with 1.4-fold within individual tasks. To quantify the energy implications, we extend an existing analytical inference-energy model to account for cached-context processing, yielding a cache-aware hybrid estimator. It estimates approximately 250 kWh for the workload, compared with approximately 10 kWh from the output-based estimator EcoLogits, a 25-fold difference largely driven by cached-context reads. Neither estimator has been validated on long cached contexts, but no modeling assumption we varied closes the gap. Starting each task from a fresh context would lower our estimate by about 60%. We also find that parent-session logs report only a subagent's final internal call, undercounting delegated tokens by a factor of ten; for this workload the correction changes the energy estimate by about 5%, but token-volume studies are fully exposed to it. Our findings expose a substantial gap between how current energy estimators and agentic workloads treat context, highlighting the need for workflow-level energy accounting that captures context accumulation, cache reuse, and delegated execution.

cs.SE↗

A Statistical Framework for Algorithmic Collective Action with Multiple Collectives

As learning systems increasingly shape everyday decisions, Algorithmic Collective Action (ACA), i.e., users coordinating changes to shared data to steer model behavior, offers a complement to regulator-side policy and corporate model design. Real-world collective actions have traditionally been decentralized and fragmented into multiple collectives, despite sharing overarching objectives, with each collective differing in size, strategy, and actionable goals. However, most of the ACA literature focuses on single collective settings. To address this, we propose the first comprehensive statistical framework for ACA with multiple collectives acting on the same system. In particular, we focus on collective action in classification, studying how multiple collectives can influence a classifier's behavior. We provide quantitative statistical bounds on the success of the collectives, considering the role and the interplay of the collectives' sizes and the alignment of their goals. We make such bounds computable by each collective with only partial knowledge of other collectives' sizes and strategies. Finally, we numerically illustrate our framework on simulations inspired by interventions for climate adaptation in smart cities, demonstrating the usefulness of our bounds.

stat.ME↗

Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves

Modern deep learning architectures increasingly contend with sophisticated signals that are natively infinite-dimensional, such as time series, probability distributions, or operators, and are defined over irregular domains. Yet, a unified learning theory for these settings has been lacking. To start addressing this gap, we introduce a novel convolutional learning framework for possibly infinite-dimensional signals supported on a manifold. Namely, we use the connection Laplacian associated with a Hilbert bundle as a convolutional operator, and we derive filters and neural networks, dubbed as \textit{HilbNets}. We make HilbNets and, more generally, the convolution operation, implementable via a two-stage sampling procedure. First, we show that sampling the manifold induces a Hilbert Cellular Sheaf, a generalized graph structure with Hilbert feature spaces and edge-wise coupling rules, and we prove that its sheaf Laplacian converges in probability to the underlying connection Laplacian as the sampling density increases. Notably, this result is a generalization to the infinite-dimensional bundle setting of the Belkin \& Niyogi \cite{BELKIN20081289} convergence result for the graph Laplacian to the manifold Laplacian, a theoretical cornerstone of geometric learning methods. Second, we discretize the signals and prove that the discretized (implementable) HilbNets converge to the underlying continuous architectures and are transferable across different samplings of the same bundle, providing consistency for learning. Finally, we validate our framework on synthetic and real-world tasks. Overall, our results broaden the scope of geometric learning as a whole by lifting classical Laplacian-based frameworks to settings where the signal at each point lives in its own Hilbert space.

cs.LG↗

LocaLeJEPA: Maximum Entropy Representations under Geometric Priors

Joint-Embedding Predictive Architectures (JEPA) offer an appealing approach to self-supervised learning, but are typically prone to representation collapse. LeJEPA resolves this by introducing a theoretically grounded distribution-matching objective that drives the embedding towards an isotropic Gaussian. However, this approach is fundamentally tied to Euclidean assumptions, and its global target does not reflect domain-specific geometric priors, such as the use of non-Euclidean embedding geometries, e.g., equivariant CNNs or hyperbolic ViTs. We thus introduce LocaLeJEPA, a framework motivated by a manifold-theoretic reformulation of LeJEPA's entropy-maximization property. The accompanying LocaLeReg, a novel locally Euclidean distribution-matching regularizer, enables a stable and scalable implementation of the LocaLeJEPA framework. We then perform extensive experiments in the vision setting on models trained on ImageNet-1K. LocaLeJEPA significantly outperforms LeJEPA when paired with hyperbolic ViT-S/8 and equivariant ResNet-50 backbones, and also with ordinary Euclidean ViT-S/8 and ViT-L/14 backbones. This holds both in terms of linear and k-NN classification performance on ImageNet and on a series of downstream tasks tested. We view the LocaLeJEPA formalism as a principled approach to the JEPA paradigm that better respects intrinsic data geometry while also accommodating generalizations to exotic embedding geometries. Code available at https://github.com/alessioborgi/LocaLeJEPA.

cs.LG↗

Ambient Discrete Diffusion: Using the Wrong Data at the Right Time for Data Efficient Learning

We introduce RefineMix, a framework for training discrete diffusion models under severe data scarcity, a common constraint in scientific applications. RefineMix uses out-of-distribution data at selected diffusion times to improve generalization without biasing the sampling distribution. Although this strategy has been explored in continuous diffusion, discrete diffusion presents a distinct challenge: unlike Gaussian noise, masking preserves domain information in surviving tokens, limiting the use of related data at high noise levels. At low noise levels, however, the domains effectively disjoint supports become an advantage, allowing the model to learn from both in-domain and out-of-distribution data without biasing the sampler. We formalize these intuitions and provide a theoretical analysis for the proposed method. Experimentally, across five domain-shift settings, RefineMix matches or outperforms in-domain finetuning and data mixing. For protein sequence generation, finetuning with just 197 in-domain examples nearly doubles the fraction of generated proteins that are simultaneously novel, foldable, and in-family compared to standard finetuning.

cs.LG↗

Noise Emissions from Hyperscale AI Data Centers: A Poorly Characterized Exposure at the Doorstep of Community Health Research

The rapid build-out of hyperscale computing for artificial-intelligence workloads has made data-center noise a recurring source of community concern in the United States, yet calibrated, standards-grade acoustic data remain scarce. This narrative review synthesizes direct measurements, regulatory and legal records, investigative journalism, and relevant engineering literature to assess what is currently known and identify the measurements needed for credible health inference. Available sources report heterogeneous acoustic values at residences, property lines, and near-source positions, with individual readings from 39 to above 70 dB(A). Because location, metric, duration, instrumentation, and provenance differ, these values are not directly comparable community exposures, do not define a pooled distribution, and do not support an aggregate comparison with nighttime limits. Among cases with spectral detail, all but one reported continuous tonal noise associated with complaints, most often low-frequency content near 70-140 Hz, that A-weighted metrics alone may under-represent. Engineering evidence further suggests that cooling architecture may be a major determinant of facility sound emission; the cited 15-30 dB span is an equipment-level sound-power estimate, not a measured facility- or residential-receptor difference, and remains a hypothesis pending comparative field measurements. We propose a standards-based field-measurement and modeling protocol designed to generate frequency-resolved exposure data at named facilities. Such exposure characterization is a necessary foundation for credible assessment of community health effects and environmental justice implications.

stat.AP↗

From Tokens to Watt-hours: Analytical Energy Estimation for LLM Inference on Modern GPUs

The operational energy consumption of large language model (LLM) inference is becoming an increasingly important component of the environmental footprint of deployed AI systems. However, direct measurement of inference energy often requires hardware telemetry, power instrumentation, or infrastructure-specific monitoring, limiting its applicability in comparative studies, early-stage system design, and sustainability reporting. This report presents an analytically structured, empirically calibrated, GPU-level methodology for estimating LLM inference energy on NVIDIA H100-class accelerators without direct runtime measurement. The proposed estimator combines parameter-scaled transformer FLOP accounting, calibrated memory-traffic factors, and hardware-specific energy coefficients for FP16/BF16 tensor-core computation and high-bandwidth-memory movement. It explicitly separates prompt prefill from autoregressive decoding, enabling energy estimates for input tokens, output tokens, and complete inference requests. The methodology further decomposes total energy into compute, parameter-access, key-value-cache write, and attention-read components, allowing the scaling behavior with model size, context length, and generated-token count to be analyzed. The resulting estimates are not intended to replace physical power measurements; rather, they provide transparent, reproducible, and assumption-explicit approximations suitable for model comparison, green-coding analysis, and design-time evaluation of LLM inference workloads.

cs.LG↗

The Hidden Water Geography of U.S. Hyperscale Data Centers in the AI Era

Water use by data centers is routinely reported as a single footprint, but water is consumed through two physically distinct pathways: at the site for cooling and in the power system that generates electricity. We mapped both pathways for 472 U.S. hyperscale facilities by linking facility locations to electricity regions, hydrologic basins, and water-stress data. Under baseline assumptions, operational water consumption totals approximately 300 GL yr^-1 (range 205-451 across scenarios), with electricity-related water contributing three-quarters of the total. The two pathways produce different hotspot geographies: direct cooling burdens concentrate in stressed western and south-central basins, whereas electricity-related burdens concentrate in a few eastern grid regions with fossil-heavy supply. Just 3 of 24 hosting balancing authorities account for 59% of electricity-related water. Separating pathways identifies which decisions matter where: cooling design and water sourcing locally, electricity planning and procurement regionally

cs.CY↗

Assessing the Carbon Emissions and Energy Consumption of U.S. Hyperscale Data Centers

The rapid proliferation of hyperscale data centers (HDCs) in the US, mainly driven by the adoption of artificial intelligence, has raised concerns about this industry's environmental footprint. We compiled facility-level information on 403 US hyperscale data centers operating between May 2024 and April 2025 and estimated their electricity consumption, electricity sources, and attributable CO2 emissions. Across different facility-load scenarios, these HDCs consumed approximately 68-99 TWh of electricity and were associated with about 37-54 million metric tons of CO2. Under the central scenario, HDC electricity demand corresponded to approximately 1.8% of total US electricity consumption, with roughly 54% of attributed generation supplied by fossil-fuel sources. The HDC electricity-weighted average carbon intensity was approximately 545 gCO2/kWh, about 48% above the contemporaneous US national grid-average carbon intensity of 370 gCO2/kWh. Our approach provides an attributional tool for assessing the environmental footprint of hyperscale data centers using the most recent EPA eGRID plant-level data.

cs.AI↗

Guided Transfer Learning for Discrete Diffusion Models

Discrete diffusion models (DMs) have achieved strong performance in language and other discrete domains, offering a compelling alternative to autoregressive modeling. Yet this performance typically depends on large training datasets, challenging the performance of DMs in small-data regimes -- common under real-world constraints. Aimed at this challenge, recent work in continuous DMs suggests that transfer learning via classifier ratio-based guidance can adapt a pretrained DM to a related target distribution, often outperforming alternatives such as full-weight fine-tuning on the target data. By contrast, transfer learning for discrete DMs remains unexplored. We address this gap by exploring practical analogues of ratio-based transfer learning for discrete DMs. Our theoretical analysis shows that a direct extension of existing ratio-based guidance is computationally prohibitive, scaling with vocabulary size. To overcome this limitation, we introduce a scheduling mechanism that yields a practical algorithm, Guided Transfer Learning for discrete diffusion models (GTL). GTL enables sampling from a target distribution without modifying the pretrained denoiser and reduces the cost to linear scaling in vocabulary size, which in turn supports longer sequence generation. We evaluate GTL on sequential data, including synthetic Markov chains and language modeling tasks, and provide a detailed empirical analysis of its behavior. The results highlight a clear trade-off: when target datasets are large, weight fine-tuning is often preferable, whereas GTL becomes increasingly effective as target data shrinks. Finally, we experimentally demonstrate a key failure mode of GTL: when the source and target distributions overlap poorly, the ratio-based classifier required for guidance becomes unreliable, limiting transfer performance.

cs.LG↗

Bayesian Nonparametrics for Principal Stratification with Continuous Post-Treatment Variables

Principal stratification provides a causal inference framework for investigating treatment effects in the presence of a post-treatment variable. Principal strata play a key role in characterizing the treatment effect by identifying groups of units with the same or similar values for the potential post-treatment variable at all treatment levels. The literature has focused mainly on binary post-treatment variables. Few papers considered continuous post-treatment variables. In the presence of a continuous post-treatment, a challenge is how to identify and characterize meaningful coarsening of the latent principal strata that lead to interpretable principal causal effects. This paper introduces the Confounders-Aware SHared atoms BAyesian mixture (CASBAH), a novel approach for principal stratification with binary treatment and continuous post-treatment variables. CASBAH leverages Bayesian nonparametric priors with an innovative hierarchical structure for the potential post-treatment outcomes that overcomes some of the limitations of previous works. Specifically, the novel features of our method allow for (i) identifying coarsened principal strata through a data-adaptive approach and (ii) providing a comprehensive quantification of the uncertainty surrounding stratum membership. Through Monte Carlo simulations, we show that the proposed methodology performs better than existing methods in characterizing the principal strata and estimating principal effects of the treatment. Finally, CASBAH is applied to a case study in which we estimate the causal effects of US national air quality regulations on pollution levels and health outcomes.

stat.ME↗

Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding

Graph Neural Networks (GNNs) excel at learning from pairwise interactions but often overlook multi-way and hierarchical relationships. Topological Deep Learning (TDL) addresses this limitation by leveraging combinatorial topological spaces. However, existing TDL models are restricted to undirected settings and fail to capture the higher-order directed patterns prevalent in many complex systems, e.g., brain networks, where such interactions are both abundant and functionally significant. To fill this gap, we introduce Semi-Simplicial Neural Networks (SSNs), a principled class of TDL models that operate on semi-simplicial sets -- combinatorial structures that encode directed higher-order motifs and their directional relationships. To enhance scalability, we propose Routing-SSNs, which dynamically select the most informative relations in a learnable manner. We prove that SSNs are strictly more expressive than standard graph and TDL models. We then introduce a new principled framework for brain dynamics representation learning, grounded in the ability of SSNs to provably recover topological descriptors shown to successfully characterize brain activity. Empirically, SSNs achieve state-of-the-art performance on brain dynamics classification tasks, outperforming the second-best model by up to 27%, and message passing GNNs by up to 50% in accuracy. Our results highlight the potential of principled topological models for learning from structured brain data, establishing a unique real-world case study for TDL. We also test SSNs on standard node classification and edge regression tasks, showing competitive performance. We will make the code and data publicly available.

cs.LG↗

Long-term impact of PM2.5 on mortality is exacerbated when wildfire events occur

There is extensive evidence that long-term exposure to all-source PM2.5 increases mortality. However, to date, no study has evaluated whether this effect is exacerbated in the presence of wildfire events. Here, we study 60+ million older US adults and find that wildfire events increase the harmful effects of long-term all-source PM2.5 exposure on mortality, providing a new and realistic conceptualization of wildfire health risks.

q-bio.PE↗

Understanding Spatial Regression Models from a Weighting Perspective in an Observational Study of Superfund Remediation

A key challenge in environmental health research is unmeasured spatial confounding, driven by unobserved spatially structured variables that influence both treatment and outcome. A common approach is to fit a spatial regression that models the outcome as a linear function of treatment and covariates, with a spatially structured error term to account for unmeasured spatial confounding. However, it remains unclear to what extent spatial regression actually accounts for such forms of confounding in finite samples, and whether this regression adjustment can be reformulated from a design-based perspective. Motivated by an observational study on the effect of Superfund site remediation on birth outcomes, we present a weighting framework for causal inference that unifies three canonical classes of spatial regression models$\unicode{x2013}$random effects, conditional autoregressive, and Gaussian process models$\unicode{x2013}$and reveals how they implicitly construct causal contrasts across space. Specifically, we show that: (i) the spatial error term induces approximate balance on a latent set of covariates and therefore adjusts for a specific form of unmeasured confounding; and (ii) the covariance structure of the spatial error can be equivalently represented as regressors in a linear model. Building on these insights, we introduce a new estimator that jointly addresses multiple forms of unmeasured spatial confounding and develop visual diagnostics. Using our new estimator, we find evidence of a small but beneficial effect of remediation on the percentage of small vulnerable newborns.

stat.ME↗

An Instrumental Variables Framework to Unite Spatial Confounding Methods

Studies investigating the causal effects of spatially varying exposures on outcomes often rely on observational and spatially indexed data. A prevalent challenge is unmeasured spatial confounding, where an unobserved spatially varying variable affects both exposure and outcome, leading to biased estimates and invalid confidence intervals. There is a very large literature on spatial statistics that attempts to address unmeasured spatial confounding bias; most of this literature is not framed in the context of causal inference and relies on strict assumptions. In this paper, we propose an instrumental variables (IV) framework that unifies and extends existing methods for addressing unmeasured spatial confounding bias. This framework reveals that many spatial confounding methods can be viewed as IV methods, in which small-scale spatial variation in exposure operates as the instrument, providing a common theoretical foundation for approaches that previously appeared distinct. The framework clarifies that these methods share a common set of assumptions and differ primarily in how small-scale spatial variation is defined, while offering a general strategy for constructing instruments. It also extends to identify and estimate a broad class of causal effects, including the exposure response curve, without requiring a linear outcome model. We apply our methodology in simulation and to a national data set of 33,255 zip codes to estimate the effect of enforcing air pollution exposure levels below 6-12 $μg/m^3$ on all-cause mortality while adjusting for unmeasured spatial confounding.

stat.ME↗

Characterizing the Effects of Environmental Exposures on Social Mobility: Bayesian Semi-parametrics for Principal Stratification

Understanding the causal effects of air pollution exposures on social mobility is attracting increasing attention. At the same time, education is widely recognized as a key driver of social mobility. However, the causal pathways linking fine particulate matter (PM2.5) exposure, educational attainment, and social mobility remain largely unexplored. To address this, we adopt the principal stratification approach, which rigorously defines causal effects when a post-treatment variable--educational attainment--is affected by exposure--PM2.5--and may, in turn, affect the primary outcome--social mobility. To estimate the causal effects, we propose a Bayesian semi-parametric method leveraging infinite mixtures for modeling the primary outcome. The proposed method (i) allows flexible modeling of the distribution of the primary potential outcomes, (ii) improves the accuracy of counterfactual imputation--a fundamental problem in causal inference framework--, and (iii) enables the characterization of treatment effects across different values of the post-treatment variable. We evaluate the performance of the proposed methodology through a Monte Carlo simulation study, demonstrating its advantages over existing approaches. Finally, we apply our method to a national dataset of 3,009 counties in the United States to estimate the causal effect of PM2.5 on social mobility, taking into account educational attainment as a post-treatment variable. Our findings indicate that in counties where higher PM2.5 exposure significantly reduces educational attainment social mobility decreases by approximately 5% compared to counties with lower PM2.5 exposure. We also find that in counties where exposure to PM2.5 does not affect educational attainment, social mobility is reduced by approximately 2% hinting at the possibility of further, yet unexplored, pathways connecting air pollution and social mobility.

stat.ME↗

Algorithmic Collective Action with Multiple Collectives

As learning systems increasingly influence everyday decisions, user-side steering via Algorithmic Collective Action (ACA)-coordinated changes to shared data-offers a complement to regulator-side policy and firm-side model design. Although real-world actions have been traditionally decentralized and fragmented into multiple collectives despite sharing overarching objectives-with each collective differing in size, strategy, and actionable goals, most of the ACA literature focused on single collective settings. In this work, we present the first theoretical framework for ACA with multiple collectives acting on the same system. In particular, we focus on collective action in classification, studying how multiple collectives can plant signals, i.e., bias a classifier to learn an association between an altered version of the features and a chosen, possibly overlapping, set of target classes. We provide quantitative results about the role and the interplay of collectives' sizes and their alignment of goals. Our framework, by also complementing previous empirical results, opens a path for a holistic treatment of ACA with multiple collectives.

cs.AI↗

Quantifying Diagnostic Signal Decay in Dementia: A National Study of Medicare Hospitalization Data

Background: Artificial intelligence (AI) models in healthcare depend on the fidelity of diagnostic data, yet the quality of such data is often compromised by variability in clinical documentation practices. In dementia, a condition already prone to diagnostic ambiguity, this variability may introduce systematic distortion into claims-based research and AI model development. Methods: We analyzed Medicare Part A hospitalization data from 2016-2018 to examine patterns of dementia-related ICD-10 code utilization across more than 3,000 U.S. counties. Using a clinically informed classification of 17 ICD-10 codes grouped into five diagnostic categories, we applied the transitive Sequential Pattern Mining (tSPM+) algorithm to model temporal usage structures. We then used matrix similarity methods to compare local diagnostic patterns to national norms and fit multivariable linear regressions to identify county-level demographic and structural correlates of divergence. Findings: We found substantial geographic and demographic variation in dementia-related diagnostic code usage. Non-specific codes were dominant nationwide, while Alzheimer's disease and vascular dementia codes showed pronounced variability. Temporal sequence analysis revealed consistent transitions from specific to non-specific codes, which suggest degradation of diagnostic specificity over time. Counties with higher proportions of rural residents, Medicaid-eligible patients, and Black or Hispanic dementia patients demonstrated significantly lower similarity to national usage patterns. Our model explained 38% of the variation in local-to-national diagnostic alignment.

stat.ME↗