arXiv subjects

Mahesh K. Marina

Publications and source records attributed to Mahesh K. Marina.

16 recordsLinked to original sources

arXiv2026

Weaver: A System for AI-RAN Compute Sharing with Foundation Model Training

The emergence of AI-RAN infrastructure, which equips cell sites with GPU-accelerated hardware, creates an opportunity to colocate non-RAN workloads with primary RAN processing. We explore using this spare capacity for decentralized training of foundation models (FMs), one of the most compute-intensive AI workloads. We present the first characterization of spare GPU capacity in AI-RAN systems at both micro-scale--across transmission slots within a cell site--and macro-scale--across sites. Our analysis finds that 40-85% of GPU capacity is unused; although this capacity is temporally bursty at individual sites, it is spatially complementary across sites. To safely and efficiently harness these resources, we present Weaver, a system that opportunistically trains FMs alongside latency-critical RAN workloads without degrading RAN performance. Weaver adopts a RAN-first design: a spare-compute controller integrated into the MAC scheduler uses compute-aware scheduling to smooth RAN GPU demand and exposes more usable spare GPU capacity. A two-level elastic training framework then adapts to dynamic, heterogeneous spare capacity within and across sites. Experiments on an O-RAN-aligned system prototype show that Weaver creates up to 4.9x more usable spare compute and utilizes up to 83% of the available spare capacity. On a multi-site testbed, Weaver improves training throughput by 2.1-3.7x over baseline approaches.

cs.DC↗

arXiv2026

On Harnessing Idle Compute at the Edge for Foundation Model Training

The foundation-model ecosystem remains highly centralized because training requires immense compute resources and is therefore largely limited to large cloud operators. Edge-assisted foundation model training that harnesses spare compute on edge devices offers a more democratized alternative. However, existing edge-training approaches fall short: they struggle to match cloud-training performance, scale to larger models, fit within device memory limits, or keep communication overhead manageable. They also do not handle device heterogeneity and churn satisfactorily. We introduce Cleave, built on a structural insight: each GEMM has an asymmetric I/O pattern -- its input matrices, sent over downlink, are much larger than the partial output blocks returned over uplink -- matching edge networks where downlink bandwidth exceeds uplink by 2--10x. Exploiting this alignment with a parameter-server-centric architecture, Cleave makes per-device communication \emph{decrease} as more devices join, rather than stay constant as in conventional TP. Decomposing training into independent sub-GEMM tasks yields one scheduling abstraction that unifies memory constraints, communication overhead, and fault tolerance under device churn. Our evaluation shows that Cleave achieves cloud-comparable GPU training performance and outperforms state-of-the-art edge-training methods by 4--10x in per-batch runtime at the same device counts. Beyond this shared operating range, Cleave scales to thousands of heterogeneous devices -- a regime where prior edge-training systems cannot operate -- and achieves at least 100x faster recovery from device failures.

cs.DC↗

arXiv2026

AltTS: A Dual-Path Framework with Alternating Optimization for Multivariate Time Series Forecasting

Multivariate time series forecasting involves two qualitatively distinct factors: (i) stable within-series autoregressive (AR) dynamics, and (ii) intermittent cross-dimension interactions that can become spurious over long horizons. We argue that fitting a single model to capture both effects creates an optimization conflict: the high-variance updates needed for cross-dimension modeling can corrupt the gradients that support autoregression, resulting in brittle training and degraded long-horizon accuracy. To address this, we propose ALTTS, a dual-path framework that explicitly decouples autoregression and cross-relation (CR) modeling. In ALTTS, the AR path is instantiated with a linear predictor, while the CR path uses a Transformer equipped with Cross-Relation Self-Attention (CRSA); the two branches are coordinated via alternating optimization to isolate gradient noise and reduce cross-block interference. Extensive experiments on multiple benchmarks show that ALTTS consistently outperforms prior methods, with the most pronounced improvements on long-horizon forecasting. Overall, our results suggest that carefully designed optimization strategies, rather than ever more complex architectures, can be a key driver of progress in multivariate time series forecasting.

cs.LG↗

arXiv2026

TUBO: A Tailored ML Framework for Reliable Network Traffic Forecasting

Traffic forecasting based network operation optimization and management offers enormous promise but also presents significant challenges from traffic forecasting perspective. While deep learning models have proven to be relatively more effective than traditional statistical methods for time series forecasting, their reliability is not satisfactory due to their inability to effectively handle unique characteristics of network traffic. In particular, the burst and complex traffic patterns makes the existing models less reliable, as each type of deep learning model has limited capability in capturing traffic patterns. To address this issue, we introduce TUBO, a novel machine learning framework custom designed for reliable network traffic forecasting. TUBO features two key components: burst processing for handling significant traffic fluctuations and model selection for adapting to varying traffic patterns using a pool of models. A standout feature of TUBO is its ability to provide deterministic predictions along with quantified uncertainty, which serves as a cue for identifying the most reliable forecasts. Evaluations on three real-world network demand matrix (DM) datasets (Abilene, GEANT, and CERNET) show that TUBO significantly outperforms existing methods on forecasting accuracy (by 4 times), and also achieves up to 94% accuracy in burst occurrence forecasting. Furthermore, we also consider traffic demand forecasting based proactive traffic engineering (TE) as a downstream use case. Our results show that compared to reactive approaches and proactive TE using the best existing DM forecasting methods, proactive TE powered by TUBO improves aggregated throughput by 9 times and 3 times, respectively.

cs.LG↗

arXiv2025

Campus5G: A Campus Scale Private 5G Open RAN Testbed

Mobile networks are embracing disaggregation, reflected by the industry trend towards Open RAN. Private 5G networks are viewed as particularly suitable contenders as early adopters of Open RAN, owing to their setting, high degree of control, and opportunity for innovation they present. Motivated by this, we have recently deployed Campus5G, the first of its kind campus-wide, O-RAN-compliant private 5G testbed across the central campus of the University of Edinburgh. We present in detail our process developing the testbed, from planning, to architecting, to deployment, and measuring the testbed performance. We then discuss the lessons learned from building the testbed, and highlight some research opportunities that emerged from our deployment experience.

cs.NI↗

arXiv2025

Black-Box Evasion Attacks on Data-Driven Open RAN Apps: Tailored Design and Experimental Evaluation

The impending adoption of Open Radio Access Network (O-RAN) is fueling innovation in the RAN towards data-driven operation. Unlike traditional RAN where the RAN data and its usage is restricted within proprietary and monolithic RAN equipment, the O-RAN architecture opens up access to RAN data via RAN intelligent controllers (RICs), to third-party machine learning (ML) powered applications - rApps and xApps - to optimize RAN operations. Consequently, a major focus has been placed on leveraging RAN data to unlock greater efficiency gains. However, there is an increasing recognition that RAN data access to apps could become a source of vulnerability and be exploited by malicious actors. Motivated by this, we carry out a comprehensive investigation of data vulnerabilities on both xApps and rApps, respectively hosted in Near- and Non-real-time (RT) RIC components of O-RAN. We qualitatively analyse the O-RAN security mechanisms and limitations for xApps and rApps, and consider a threat model informed by this analysis. We design a viable and effective black-box evasion attack strategy targeting O-RAN RIC Apps while accounting for the stringent timing constraints and attack effectiveness. The strategy employs four key techniques: the model cloning algorithm, input-specific perturbations, universal adversarial perturbations (UAPs), and targeted UAPs. This strategy targets ML models used by both xApps and rApps within the O-RAN system, aiming to degrade network performance. We validate the effectiveness of the designed evasion attack strategy and quantify the scale of performance degradation using a real-world O-RAN testbed and emulation environments. Evaluation is conducted using the Interference Classification xApp and the Power Saving rApp as representatives for near-RT and non-RT RICs. We also show that the attack strategy is effective against prominent defense techniques for adversarial ML.

cs.CR↗

arXiv2025

BISCAY: Practical Radio KPI Driven Congestion Control for Mobile Networks

Mobile application performance relies heavily on the congestion control design of the underlying transport, which is typically bottlenecked by cellular link and has to cope with rapid cellular link bandwidth fluctuations. We observe that radio KPI measurements from the mobile device chipset can be exploited for precise and timely measurement of available bandwidth on the cellular link. Building on this insight, we propose Biscay, a practical and radio KPI-driven congestion control system design for mobile networks. Biscay leverages OpenDiag, the in-kernel real-time radio KPI extraction tool we introduce in this paper, along with our KPI-based accurate bandwidth determination layer towards dynamically adjusting the congestion window to optimally use the available bandwidth while keeping delay to the minimum. Our solution is practical and deployable, as shown through our implementation of Biscay and OpenDiag on unrooted Android 5G phones. We extensively evaluate Biscay against different state-of-the-art congestion control designs including BBR and CUBIC with emulations driven by real measurement traces as well as real-world experiments spanning diverse 4G and 5G scenarios, and show that it provides significant average and tail delay improvements (typically over 90% reduction) while yielding better or similar throughput. These gains are enabled by 100% improvement in the granularity of on-device radio KPI measurements with OpenDiag compared to existing alternatives like MobileInsight.

cs.NI↗

arXiv2025

Towards Decentralized and Sustainable Foundation Model Training with the Edge

Foundation models are at the forefront of AI research, appealing for their ability to learn from vast datasets and cater to diverse tasks. Yet, their significant computational demands raise issues of environmental impact and the risk of centralized control in their development. We put forward a vision towards decentralized and sustainable foundation model training that leverages the collective compute of sparingly used connected edge AI devices. We present the rationale behind our vision, particularly in support of its sustainability benefit. We further outline a set of challenges that need to be addressed to turn this vision into reality.

cs.LG↗

arXiv2025

HybridServe: Efficient Serving of Large AI Models with Confidence-Based Cascade Routing

Giant Deep Neural Networks (DNNs), have become indispensable for accurate and robust support of large-scale cloud based AI services. However, serving giant DNNs is prohibitively expensive from an energy consumption viewpoint easily exceeding that of training, due to the enormous scale of GPU clusters needed to hold giant DNN model partitions and replicas. Existing approaches can either optimize energy efficiency or inference accuracy but not both. To overcome this status quo, we propose HybridServe, a novel hybrid DNN model serving system that leverages multiple sized versions (small to giant) of the model to be served in tandem. Through a confidence based hybrid model serving dataflow, HybridServe prefers to serve inference requests with energy-efficient smaller models so long as accuracy is not compromised, thereby reducing the number of replicas needed for giant DNNs. HybridServe also features a dataflow planner for efficient partitioning and replication of candidate models to maximize serving system throughput. Experimental results using a prototype implementation of HybridServe show that it reduces energy footprint by up to 19.8x compared to the state-of-the-art DNN model serving systems while matching the accuracy of serving solely with giant DNNs.

cs.LG↗

arXiv2024

PH-Dropout: Practical Epistemic Uncertainty Quantification for View Synthesis

View synthesis using Neural Radiance Fields (NeRF) and Gaussian Splatting (GS) has demonstrated impressive fidelity in rendering real-world scenarios. However, practical methods for accurate and efficient epistemic Uncertainty Quantification (UQ) in view synthesis are lacking. Existing approaches for NeRF either introduce significant computational overhead (e.g., ``10x increase in training time" or ``10x repeated training") or are limited to specific uncertainty conditions or models. Notably, GS models lack any systematic approach for comprehensive epistemic UQ. This capability is crucial for improving the robustness and scalability of neural view synthesis, enabling active model updates, error estimation, and scalable ensemble modeling based on uncertainty. In this paper, we revisit NeRF and GS-based methods from a function approximation perspective, identifying key differences and connections in 3D representation learning. Building on these insights, we introduce PH-Dropout (Post hoc Dropout), the first real-time and accurate method for epistemic uncertainty estimation that operates directly on pre-trained NeRF and GS models. Extensive evaluations validate our theoretical findings and demonstrate the effectiveness of PH-Dropout.

cs.CV↗

arXiv2024

Learning High-Frequency Functions Made Easy with Sinusoidal Positional Encoding

Fourier features based positional encoding (PE) is commonly used in machine learning tasks that involve learning high-frequency features from low-dimensional inputs, such as 3D view synthesis and time series regression with neural tangent kernels. Despite their effectiveness, existing PEs require manual, empirical adjustment of crucial hyperparameters, specifically the Fourier features, tailored to each unique task. Further, PEs face challenges in efficiently learning high-frequency functions, particularly in tasks with limited data. In this paper, we introduce sinusoidal PE (SPE), designed to efficiently learn adaptive frequency features closely aligned with the true underlying function. Our experiments demonstrate that SPE, without hyperparameter tuning, consistently achieves enhanced fidelity and faster training across various tasks, including 3D view synthesis, Text-to-Speech generation, and 1D regression. SPE is implemented as a direct replacement for existing PEs. Its plug-and-play nature lets numerous tasks easily adopt and benefit from SPE.

cs.LG↗

arXiv2020

Examining Machine Learning for 5G and Beyond through an Adversarial Lens

Spurred by the recent advances in deep learning to harness rich information hidden in large volumes of data and to tackle problems that are hard to model/solve (e.g., resource allocation problems), there is currently tremendous excitement in the mobile networks domain around the transformative potential of data-driven AI/ML based network automation, control and analytics for 5G and beyond. In this article, we present a cautionary perspective on the use of AI/ML in the 5G context by highlighting the adversarial dimension spanning multiple types of ML (supervised/unsupervised/RL) and support this through three case studies. We also discuss approaches to mitigate this adversarial ML risk, offer guidelines for evaluating the robustness of ML models, and call attention to issues surrounding ML oriented research in 5G more generally.

cs.NI↗

arXiv2019

Iris: Deep Reinforcement Learning Driven Shared Spectrum Access Architecture for Indoor Neutral-Host Small Cells

We consider indoor mobile access, a vital use case for current and future mobile networks. For this key use case, we outline a vision that combines a neutral-host based shared small-cell infrastructure with a common pool of spectrum for dynamic sharing as a way forward to proliferate indoor small-cell deployments and open up the mobile operator ecosystem. Towards this vision, we focus on the challenges pertaining to managing access to shared spectrum (e.g., 3.5GHz US CBRS spectrum). We propose Iris, a practical shared spectrum access architecture for indoor neutral-host small-cells. At the core of Iris is a deep reinforcement learning based dynamic pricing mechanism that efficiently mediates access to shared spectrum for diverse operators in a way that provides incentives for operators and the neutral-host alike. We then present the Iris system architecture that embeds this dynamic pricing mechanism alongside cloud-RAN and RAN slicing design principles in a practical neutral-host design tailored for the indoor small-cell environment. Using a prototype implementation of the Iris system, we present extensive experimental evaluation results that not only offer insight into the Iris dynamic pricing process and its superiority over alternative approaches but also demonstrate its deployment feasibility.

cs.NI↗

arXiv2019

Urban Vibes and Rural Charms: Analysis of Geographic Diversity in Mobile Service Usage at National Scale

We investigate spatial patterns in mobile service consumption that emerge at national scale. Our investigation focuses on a representative case study, i.e., France, where we find that: (i) the demand for popular mobile services is fairly uniform across the whole country, and only a reduced set of peculiar services (mainly operating system updates and long-lived video streaming) yields geographic diversity; (ii) even for such distinguishing services, the spatial heterogeneity of demands is limited, and a small set of consumption behaviors is sufficient to characterize most of the mobile service usage across the country; (iii) the spatial distribution of these behaviors correlates well with the urbanization level, ultimately suggesting that the adoption of geographically-diverse mobile applications is linked to a dichotomy of cities and rural areas. We derive our results through the analysis of substantial measurement data collected by a major mobile network operator, leveraging an approach rooted in information theory that can be readily applied to other scenarios.

cs.NI↗

arXiv2016

Holistic Small Cell Traffic Balancing across Licensed and Unlicensed Bands

Due to the dramatic growth in mobile data traffic on one hand and the scarcity of the licensed spectrum on the other hand, mobile operators are considering the use of unlicensed bands (especially those in 5 GHz) as complementary spectrum for providing higher system capacity and better user experience. This approach is currently being standardized by 3GPP under the name of LTE Licensed-Assisted Access (LTE-LAA). In this paper, we take a holistic approach for LTE-LAA small cell traffic balancing by jointly optimizing the use of the licensed and unlicensed bands. We pose this traffic balancing as an optimization problem that seeks proportional fair coexistence of WiFi, small cell and macro cell users by adapting the transmission probability of the LTE-LAA small cell in the licensed and unlicensed bands. The motivation for this formulation is for the LTE-LAA small cell to switch between or aggregate licensed and unlicensed bands depending on the interference/traffic level and the number of active users in each band. We derive a closed form solution for this optimization problem and additionally propose a transmission mechanism for the operation of the LTE-LAA small cell on both bands. Through numerical and simulation results, we show that our proposed traffic balancing scheme, besides enabling better LTE-WiFi coexistence and efficient utilization of the radio resources relative to the existing traffic balancing scheme, also provides a better tradeoff between maximizing the total network throughput and achieving fairness among all network flows compared to alternative approaches.

cs.NI↗

arXiv2013

Auctioning based Coordinated TV White Space Spectrum Sharing for Home Networks

The idea of having the geolocation database monitor the secondary use of TV white space (TVWS) spectrum and assist in coordinating the secondary usage is gaining ground. Considering the home networking use case, we leverage the geolocation database for interference-aware coordinated TVWS sharing among secondary users (home networks) using {\em short-term auctions}, thereby realize a dynamic secondary market. To enable this auctioning based coordinated TVWS sharing framework, we propose an enhanced {\em market-driven TVWS spectrum access model}. For the short-term auctions, we propose an online multi-unit, iterative truthful mechanism called VERUM that takes into consideration spatially heterogeneous spectrum availability, an inherent characteristic in the TVWS context. We prove that VERUM is truthful (i.e., the best strategy for every bidder is to bid based on its true valuation) and is also efficient in that it allocates spectrum to users who value it the most. Evaluation results from scenarios with real home distributions in urban and dense-urban environments and using realistic TVWS spectrum availability maps show that VERUM performs close to optimal allocation in terms of revenue for the coordinating spectrum manager. Comparison with two existing efficient and truthful multi-unit spectrum auction schemes, VERITAS and SATYA, shows that VERUM fares better in terms of revenue, spectrum utilisation and percentage of winning bidders in diverse conditions. Taking all of the above together, VERUM can be seen to offer incentives to subscribed users encouraging them to use TVWS spectrum through greater spectrum availability (as measured by percentage of winning bidders) as well as to the coordinating spectrum manager through revenue generation.

cs.NI↗