arXiv subjects

Gustavo de Veciana

Publications and source records attributed to Gustavo de Veciana.

At least 19 recordsLinked to original sources

arXiv2026

Optimal Resource Allocation for ML Model Training and Deployment under Concept Drift

We study how to allocate resources for training and deployment of machine learning (ML) models under concept drift and limited budgets. We consider a setting in which a model provider distributes trained models to multiple clients whose devices support local inference but lack the ability to retrain those models, placing the burden of performance maintenance on the provider. We introduce a model-agnostic framework that captures the interaction between resource allocation, concept drift dynamics, and deployment timing. We show that optimal training policies depend critically on the aging properties of concept durations. Under sudden concept changes, we derive optimal training policies subject to budget constraints when concept durations follow distributions with Decreasing Mean Residual Life (DMRL), and show that intuitive heuristics are provably suboptimal under Increasing Mean Residual Life (IMRL). We further study model deployment under communication constraints, prove that the associated optimization problem is quasi-convex under mild conditions, and propose a randomized scheduling strategy that achieves near-optimal client-side performance. These results offer theoretical and algorithmic foundations for cost-efficient ML model management under concept drift, with implications for continual learning, distributed inference, and adaptive ML systems.

cs.LG↗

arXiv2026

Efficient Weighted Sampling via Score-based Generative Models

Weighted sampling -- sampling from a probability density function (PDF) proportional to the product of a base PDF and a weight function -- is a fundamental technique with wide-ranging applications in variance reduction, biased sampling, data augmentation, and more. Leveraging the increasing availability of pretrained score-based generative models (SGMs), we propose a training-free weighted sampling framework that approximates the backward diffusion process of the target distribution by augmenting the pretrained base score function with an auxiliary guidance term, in a principled and computationally efficient manner. Our approach builds on two key components: a lightweight approximation of the guidance that avoids costly higher-order derivatives of both the score and weight functions, and an uncertainty-aware scheduler that dynamically adjusts the guidance strength based on a temporal analysis of approximation error. Together, these components enable accurate and stable sampling without relying on particle-based resampling or Hessian evaluations commonly required by existing methods. We validate the effectiveness of our method from synthetic to large-scale settings such as Stable Diffusion XL, where our framework achieves $1.2\times$ to $4.7\times$ speedups while consistently matching or outperforming state-of-the-art baselines in task performance. These results position our method as a scalable and inference-efficient solution for task-adaptive, time-sensitive sampling in generative applications.

cs.LG↗

arXiv2026

CoreQ: Learning-Free Mismatch Correction and Successive Rounding for Quantization

Post-training quantization (PTQ) enables efficient deployment of large language models by mapping pretrained weights to low-bit formats without retraining, typically using a small calibration set to minimize a layer-wise calibration objective. However, this sequential procedure induces a mismatch: errors from earlier quantized layers alter the inputs received by later layers, causing the activations to deviate from those of the full-precision model. Recent approaches introduce mismatch-aware calibration objectives to compensate for this effect, but leave open how much of the observed mismatch should shift each layer's calibration target. Fully applying this correction can overfit limited calibration data, while scaling the mismatch correction with a fixed coefficient ignores varying reliability of mismatch estimates across layers. To address these limitations, we propose CoreQ, a learning-free PTQ framework that applies a closed-form coefficient for mismatch correction derived from a geometric decomposition of the mismatch. The resulting coefficient adapts the correction across layers, reduces overfitting to finite calibration data, and requires no hyperparameter tuning. Given the corrected target, CoreQ minimizes the induced triangular least-squares objective with an efficient greedy successive-rounding solver and a bounded beam-search extension, K-CoreQ, that trades modest additional compute for improved performance. Across multiple LLM families, scales, bit-widths, and quantization settings, CoreQ improves perplexity and downstream accuracy over strong PTQ baselines.

cs.LG↗

arXiv2026

Online Learning for Multi-Layer Hierarchical Inference under Partial and Policy-Dependent Feedback

Hierarchical inference systems route tasks across multiple computational layers, where each node may either finalize a prediction locally or offload the task to a node in the next layer for further processing. Learning optimal routing policies in such systems is challenging: inference loss is defined recursively across layers, while feedback on prediction error is revealed only at a terminal oracle layer. This induces a partial, policy-dependent feedback structure in which observability probabilities decay with depth, causing importance-weighted estimators to suffer from amplified variance. We study online routing for multi-layer hierarchical inference under long-term resource constraints and terminal-only feedback. We formalize the recursive loss structure and show that naive importance-weighted contextual bandit methods become unstable as feedback probability decays along the hierarchy. To address this, we develop a variance-reduced EXP4-based algorithm integrated with Lyapunov optimization, yielding unbiased loss estimation and stable learning under sparse and policy-dependent feedback. We provide regret guarantees relative to the best fixed routing policy in hindsight and establish near-optimality under stochastic arrivals and resource constraints. Experiments on large-scale multi-task workloads demonstrate improved stability and performance compared to standard importance-weighted approaches.

cs.LG↗

arXiv2025

Inferring Causal Relationships to Improve Caching for Clients with Correlated Requests: Applications to VR

Efficient edge caching reduces latency and alleviates backhaul congestion in modern networks. Traditional caching policies, such as Least Recently Used (LRU) and Least Frequently Used (LFU), perform well under specific request patterns. LRU excels in workloads with strong temporal locality, while LFU is effective when content popularity remains static. However, real-world client requests often exhibit correlations due to shared contexts and coordinated activities. This is particularly evident in Virtual Reality (VR) environments, where groups of clients navigate shared virtual spaces, leading to correlated content requests. In this paper, we introduce the \textit{grouped client request model}, a generalization of the Independent Reference Model that explicitly captures different types of request correlations. Our theoretical analysis of LRU under this model reveals that the optimal causal caching policy depends on cache size: LFU is optimal for small to moderate caches, while LRU outperforms it for larger caches. To address the limitations of existing policies, we propose Least Following and Recently Used (LFRU), a novel online caching policy that dynamically infers and adapts to causal relationships in client requests to optimize evictions. LFRU prioritizes objects likely to be requested based on inferred dependencies, achieving near-optimal performance compared to the offline optimal Belady policy in structured correlation settings. We develop VR based datasets to evaluate caching policies under realistic correlated requests. Our results show that LFRU consistently performs at least as well as LRU and LFU, outperforming LRU by up to 2.9x and LFU by up to1.9x in certain settings.

cs.NI↗

arXiv2025

Optimal Scheduling Algorithms for LLM Inference: Theory and Practice

With the growing use of Large Language Model (LLM)-based tools like ChatGPT, Perplexity, and Gemini across industries, there is a rising need for efficient LLM inference systems. These systems handle requests with a unique two-phase computation structure: a prefill-phase that processes the full input prompt and a decode-phase that autoregressively generates tokens one at a time. This structure calls for new strategies for routing and scheduling requests. In this paper, we take a comprehensive approach to this challenge by developing a theoretical framework that models routing and scheduling in LLM inference systems. We identify two key design principles-optimal tiling and dynamic resource allocation-that are essential for achieving high throughput. Guided by these principles, we propose the Resource-Aware Dynamic (RAD) scheduler and prove that it achieves throughput optimality under mild conditions. To address practical Service Level Objectives (SLOs) such as serving requests with different Time Between Token (TBT) constraints, we design the SLO-Aware LLM Inference (SLAI) scheduler. SLAI uses real-time measurements to prioritize decode requests that are close to missing their TBT deadlines and reorders prefill requests based on known prompt lengths to further reduce the Time To First Token (TTFT) delays. We evaluate SLAI on the Openchat ShareGPT4 dataset using the Mistral-7B model on an NVIDIA RTX ADA 6000 GPU. Compared to Sarathi-Serve, SLAI reduces the median TTFT by 53% and increases the maximum serving capacity by 26% such that median TTFT is below 0.5 seconds, while meeting tail TBT latency constraints.

cs.LG↗

arXiv2025

Batching-Aware Joint Model Onloading and Offloading for Hierarchical Multi-Task Inference

The growing demand for intelligent services on resource-constrained edge devices has spurred the development of collaborative inference systems that distribute workloads across end devices, edge servers, and the cloud. While most existing frameworks focus on single-task, single-model scenarios, many real-world applications (e.g., autonomous driving and augmented reality) require concurrent execution of diverse tasks including detection, segmentation, and depth estimation. In this work, we propose a unified framework to jointly decide which multi-task models to deploy (onload) at clients and edge servers, and how to route queries across the hierarchy (offload) to maximize overall inference accuracy under memory, compute, and communication constraints. We formulate this as a mixed-integer program and introduce J3O (Joint Optimization of Onloading and Offloading), an alternating algorithm that (i) greedily selects models to onload via Lagrangian-relaxed submodular optimization and (ii) determines optimal offloading via constrained linear programming. We further extend J3O to account for batching at the edge, maintaining scalability under heterogeneous task loads. Experiments show J3O consistently achieves over $97\%$ of the optimal accuracy while incurring less than $15\%$ of the runtime required by the optimal solver across multi-task benchmarks.

cs.LG↗

arXiv2025

Fundamentals of Caching Layered Data objects

The effective management of large amounts of data processed or required by today's cloud or edge computing systems remains a fundamental challenge. This paper focuses on cache management for applications where data objects can be stored in layered representations. In such representations, each additional data layer enhances the "quality" of the object's version but comes with an incremental cost of memory space. This layered approach proves beneficial in various scenarios, including the delivery of zoomable maps, video coding, future Virtual Reality gaming, and layered neural network models where additional data layers improve inference accuracy. In systems where users or devices demand different versions of a data object, layered representations offer flexibility for caching policies to achieve improved hit rates. In this paper, we explore the performance of various traditionally studied caching policies, such as Belady, LRU, and LFU, both with and without layering. To this end, we develop an asymptotically accurate analytical model for Layered LRU (LLRU). We study how the performance of LLRU is impacted by factors such as the number of layers, the popularity of different objects and layers, and overheads associated with storing layered representations. For instance, we show that, for LLRU, more layers are not always beneficial and indeed performance depends in subtle ways on the popularity and size profiles of layers.

cs.NI↗

arXiv2024

Ten Ways in which Virtual Reality Differs from Video Streaming

Virtual Reality (VR) applications have a number of unique characteristics that set them apart from traditional video streaming. These characteristics have major implications on the design of VR rendering, adaptation, prefetching, caching, and transport mechanisms. This paper contrasts VR to video streaming, stored 2D video streaming in particular, and discusses how to rethink system and network support for VR.

cs.PF↗

arXiv2023

Network Adaptive Federated Learning: Congestion and Lossy Compression

In order to achieve the dual goals of privacy and learning across distributed data, Federated Learning (FL) systems rely on frequent exchanges of large files (model updates) between a set of clients and the server. As such FL systems are exposed to, or indeed the cause of, congestion across a wide set of network resources. Lossy compression can be used to reduce the size of exchanged files and associated delays, at the cost of adding noise to model updates. By judiciously adapting clients' compression to varying network congestion, an FL application can reduce wall clock training time. To that end, we propose a Network Adaptive Compression (NAC-FL) policy, which dynamically varies the client's lossy compression choices to network congestion variations. We prove, under appropriate assumptions, that NAC-FL is asymptotically optimal in terms of directly minimizing the expected wall clock training time. Further, we show via simulation that NAC-FL achieves robust performance improvements with higher gains in settings with positively correlated delays across time.

cs.LG↗

arXiv2021

Performance-Complexity Tradeoffs in Greedy Weak Submodular Maximization with Random Sampling

Many problems in signal processing and machine learning can be formalized as weak submodular optimization tasks. For such problems, a simple greedy algorithm (\textsc{Greedy}) is guaranteed to find a solution achieving the objective with a value no worse than $1-e^{-1/c}$ of the optimal, where $c$ is the multiplicative weak-submodularity constant. Due to the high cost of querying large-scale systems, the complexity of \textsc{Greedy} becomes prohibitive in contemporary applications. In this work, we study the tradeoff between performance and complexity when one resorts to random sampling strategies to reduce the query complexity of \textsc{Greedy}. Specifically, we quantify the effect of uniform sampling strategies on \textsc{Greedy}'s performance through two metrics: (i) probability of identifying an optimal subset, and (ii) suboptimality with respect to the optimal solution. The latter implies that uniform sampling strategies with a fixed sampling size achieve a non-trivial approximation factor; however, we show that with overwhelming probability, these methods fail to find the optimal subset. Our analysis shows that the failure of uniform sampling strategies with fixed sample size can be circumvented by successively increasing the size of the search space. Building upon this insight, we propose a simple progressive stochastic greedy algorithm and study its approximation guarantees. Moreover, we demonstrate effectiveness of the proposed method in dimensionality reduction applications and feature selection tasks for clustering and object tracking.

cs.DM↗

arXiv2021

Book-Ahead & Supply Management for Ridesourcing Platforms

Ridesourcing platforms recently introduced the ``schedule a ride'' service where passengers may reserve (book-ahead) a ride in advance of their trip. Reservations give platforms precise information that describes the start time and location of anticipated future trips; in turn, platforms can use this information to adjust the availability and spatial distribution of the driver supply. In this article, we propose a framework for modeling/analyzing reservations in time-varying stochastic ridesourcing systems. We consider that the driver supply is distributed over a network of geographic regions and that book-ahead rides have reach time priority over non-reserved rides. First, we propose a state-dependent admission control policy that assigns drivers to passengers; this policy ensures that the reach time service requirement would be attained for book-ahead rides. Second, given the admission control policy and reservations information in each region, we predict the ``target" number of drivers that is required (in the future) to probabilistically guarantee the reach time service requirement for stochastic non-reserved rides. Third, we propose a reactive dispatching/rebalancing mechanism that determines the adjustments to the driver supply that are needed to maintain the targets across regions. For a specific reach time quality of service, simulation results using data from Lyft rides in Manhattan exhibit how the number of idle drivers decreases with the fraction of book-ahead rides. We also observe that the non-stationary demand (ride request) rate varies significantly across time; this rapid variation further illustrates that time-dependent models are needed for operational analysis of ridesourcing systems.

eess.SY↗

arXiv2020

Modeling and Analysis of Data Harvesting Architecture based on Unmanned Aerial Vehicles

This paper explores an emerging wireless Internet-of-things (IoT) architecture based on unmanned aerial vehicles (UAVs). We consider a network where a fleet of UAVs at a fixed altitude flies on planned trajectories and IoT devices on the ground are scheduled to transmit their data to the UAVs when the latter are nearby. In such a system, the UAVs' motion triggers the uplink transmissions of the IoT devices. As a result, network performance is determined by the geometric and dynamic characteristics of the system. We propose a joint stationary model for UAVs and IoT devices and then evaluate the interference, the coverage probability, and the data rate of the typical UAV. To assess the harvesting capability of the proposed architecture, we derive a formula for the amount of data uploaded from each IoT device to a UAV. We also establish a linear relationship between the UAV coverage and the harvesting capability of the network, which provides insights into the design of the proposed harvesting scheme. In addition, we use our analytical results to numerically show that there exists a trade-off between the uploaded data and the size of the IoT scheduling window. Specifically, for a given UAV and IoT geometry, there exists an optimal scheduling window that maximizes the harvesting capability of the proposed network.

cs.IT↗

arXiv2020

Performance Analysis of RSU-based Multihomed Multilane Vehicular Networks

Motivated by the potentially high downlink traffic demands of commuters in future autonomous vehicles, we study a network architecture where vehicles use Vehicle-to-Vehicle (V2V) links to form relay network clusters, which in turn use Vehicle-to-Infrastructure (V2I) links to connect to one or more Road Side Units (RSUs). Such cluster-based multihoming offers improved performance, e.g., in coverage and per user shared rate, but depends on the penetration of V2V-V2I capable vehicles and possible blockage, by legacy vehicles, of line of sight based V2V links, such as those based on millimeter-wave and visible light technologies. This paper provides a performance analysis of a typical vehicle's connectivity and throughput on a highway in the free-flow regime, exploring its dependence on vehicle density, sensitivity to blockages, number of lanes and heterogeneity across lanes. The results show that, even with moderate vehicle densities and penetration of V2V-V2I capable vehicles, such architectures can achieve substantial improvements in connectivity and reduction in per-user rate variability as compared to V2I based networks. The typical vehicle's performance is also shown to improve considerably in the multilane highway setting as compared to a single lane road. This paper also sheds light on how the network performance is affected when vehicles can control their relative positions, by characterizing the connectivity-throughput tradeoff faced by the clusters of vehicles.

eess.SY↗

arXiv2020

Constrained Network Slicing Games: Achieving service guarantees and network efficiency

Network slicing is a key capability for next generation mobile networks. It enables one to cost effectively customize logical networks over a shared infrastructure. A critical component of network slicing is resource allocation, which needs to ensure that slices receive the resources needed to support their mobiles/services while optimizing network efficiency. In this paper, we propose a novel approach to slice-based resource allocation named Guaranteed seRvice Efficient nETwork slicing (GREET). The underlying concept is to set up a constrained resource allocation game, where (i) slices unilaterally optimize their allocations to best meet their (dynamic) customer loads, while (ii) constraints are imposed to guarantee that, if they wish so, slices receive a pre-agreed share of the network resources. The resulting game is a variation of the well-known Fisher market, where slices are provided a budget to contend for network resources (as in a traditional Fisher market), but (unlike a Fisher market) prices are constrained for some resources to provide the desired guarantees. In this way, GREET combines the advantages of a share-based approach (high efficiency by flexible sharing) and reservation-based ones (which provide guarantees by assigning a fixed amount of resources). We characterize the Nash equilibrium, best response dynamics, and propose a practical slice strategy with provable convergence properties. Extensive simulations exhibit substantial improvements over network slicing state-of-the-art benchmarks.

cs.NI↗

arXiv2019

Poly-Symmetry in Processor-Sharing Systems

We consider a system of processor-sharing queues with state-dependent service rates. These are allocated according to balanced fairness within a polymatroid capacity set. Balanced fairness is known to be both insensitive and Pareto-efficient in such systems, which ensures that the performance metrics, when computable, will provide robust insights into the real performance of the system considered. We first show that these performance metrics can be evaluated with a complexity that is polynomial in the system size if the system is partitioned into a finite number of parts, so that queues are exchangeable within each part and asymmetric across different parts. This in turn allows us to derive stochastic bounds for a larger class of systems which satisfy less restrictive symmetry assumptions. These results are applied to practical examples of tree data networks, such as backhaul networks of Internet service providers, and computer clusters.

cs.PF↗

arXiv2019

Modeling and Optimization of Human-machine Interaction Processes via the Maximum Entropy Principle

We propose a data-driven framework to enable the modeling and optimization of human-machine interaction processes, e.g., systems aimed at assisting humans in decision-making or learning, work-load allocation, and interactive advertising. This is a challenging problem for several reasons. First, humans' behavior is hard to model or infer, as it may reflect biases, long term memory, and sensitivity to sequencing, i.e., transience and exponential complexity in the length of the interaction. Second, due to the interactive nature of such processes, the machine policy used to engage with a human may bias possible data-driven inferences. Finally, in choosing machine policies that optimize interaction rewards, one must, on the one hand, avoid being overly sensitive to error/variability in the estimated human model, and on the other, being overly deterministic/predictable which may result in poor human 'engagement' in the interaction. To meet these challenges, we propose a robust approach, based on the maximum entropy principle, which iteratively estimates human behavior and optimizes the machine policy--Alternating Entropy-Reward Ascent (AREA) algorithm. We characterize AREA, in terms of its space and time complexity and convergence. We also provide an initial validation based on synthetic data generated by an established noisy nonlinear model for human decision-making.

cs.LG↗

arXiv2019

Elastic Multi-resource Network Slicing: Can Protection Lead to Improved Performance?

In order to meet the performance/privacy requirements of future data-intensive mobile applications, e.g., self-driving cars, mobile data analytics, and AR/VR, service providers are expected to draw on shared storage/computation/connectivity resources at the network "edge". To be cost-effective, a key functional requirement for such infrastructure is enabling the sharing of heterogeneous resources amongst tenants/service providers supporting spatially varying and dynamic user demands. This paper proposes a resource allocation criterion, namely, Share Constrained Slicing (SCS), for slices allocated predefined shares of the network's resources, which extends the traditional alpha-fairness criterion, by striking a balance among inter- and intra-slice fairness vs. overall efficiency. We show that SCS has several desirable properties including slice-level protection, envyfreeness, and load driven elasticity. In practice, mobile users' dynamics could make the cost of implementing SCS high, so we discuss the feasibility of using a simpler (dynamically) weighted max-min as a surrogate resource allocation scheme. For a setting with stochastic loads and elastic user requirements, we establish a sufficient condition for the stability of the associated coupled network system. Finally, and perhaps surprisingly, we show via extensive simulations that while SCS (and/or the surrogate weighted max-min allocation) provides inter-slice protection, they can achieve improved job delay and/or perceived throughput, as compared to other weighted max-min based allocation schemes whose intra-slice weight allocation is not share-constrained, e.g., traditional max-min or discriminatory processor sharing.

cs.NI↗