arXiv subjects

Jaya Prakash Champati

Publications and source records attributed to Jaya Prakash Champati.

At least 19 recordsLinked to original sources

arXiv2026

Pay for Hints, Not Answers: LLM Shepherding for Cost-Efficient Inference

Large Language Models (LLMs) deliver state-of-the-art performance on complex reasoning tasks, but their inference costs limit deployment at scale. Small Language Models (SLMs) offer dramatic cost savings yet lag substantially in accuracy. Existing approaches - routing and cascading - treat the LLM as an all-or-nothing resource: either the query bypasses the LLM entirely, or the LLM generates a complete response at full cost. We introduce LLM Shepherding, a framework that requests only a short prefix (a hint) from the LLM and provides it to SLM. This simple mechanism is surprisingly effective for math and coding tasks: even hints comprising 10-30% of the full LLM response improve SLM accuracy significantly. Shepherding generalizes both routing and cascading, and it achieves lower cost under oracle decision-making. We develop a two-stage predictor that jointly determines whether a hint is needed and how many tokens to request. On the widely-used mathematical reasoning (GSM8K, CNK12) and code generation (HumanEval, MBPP) benchmarks, Shepherding reduces costs by 42-94% relative to LLM-only inference. Compared to state-of-the-art routing and cascading baselines, shepherding delivers up to 2.8x cost reduction while matching accuracy. To our knowledge, this is the first work to exploit token-level budget control for SLM-LLM collaboration.

cs.LG↗

arXiv2026

Speculation at a Distance: Where Edge-Cloud Speculative Decoding Actually Pays Off

Speculative decoding (SD) accelerates LLM inference by $1.5$-$3$ times when the draft and target models are co-located. This has motivated a distributed variant (DSD) that places the draft model on an edge device while the target stays in the cloud. We show with closed-form inequalities that DSD's per-request latency benefit is limited under WAN edge-cloud communication. If the server can host both models, co-located SD has lower latency and communication than synchronous DSD, with the same per-output FLOPs and model-weight memory. Pipelining can make DSD competitive with co-located SD only in low-RTT regimes where the round trip is shorter than the edge drafting time window; at WAN RTTs, the cloud round trip remains too large for pipelined DSD to beat co-located SD. Against cloud autoregressive decoding, DSD can reduce latency only inside a bounded window given the target-model speed, acceptance rate, and RTT. DSD is also infeasible against closed-source APIs without a verifier-only interface. The main case for DSD appears in multi-tenant capacity. Under cross-client overlap, offloading draft compute lets a saturated cloud server sustain $(1 + γ\,t_d/t_v)$ times more concurrent clients at the same per-client rate, where $γ$ is the speculation length and $t_d, t_v$ are the per-step draft and verification times. DSD should therefore be evaluated primarily by multi-tenant capacity and server throughput, not only by single-request latency.

cs.DC↗

arXiv2026

Inference Offloading for Cost-Sensitive Binary Classification at the Edge

We focus on a binary classification problem in an edge intelligence system where false negatives are more costly than false positives. The system has a compact, locally deployed model, which is supplemented by a larger, remote model, which is accessible via the network by incurring an offloading cost. For each sample, our system first uses the locally deployed model for inference. Based on the output of the local model, the sample may be offloaded to the remote model. This work aims to understand the fundamental trade-off between classification accuracy and the offloading costs within such a hierarchical inference (HI) system. To optimise this system, we propose an online learning framework that continuously adapts a pair of thresholds on the local model's confidence scores. These thresholds determine the prediction of the local model and whether a sample is classified locally or offloaded to the remote model. We present a closed-form solution for the setting where the local model is calibrated. For the more general case of uncalibrated models, we introduce H2T2, an online two-threshold hierarchical inference policy, and prove it achieves sublinear regret. H2T2 is model-agnostic, requires no training, and learns during the inference phase using limited feedback. Simulations on real-world datasets show that H2T2 consistently outperforms naive and single-threshold HI policies, sometimes even surpassing offline optima. The policy also demonstrates robustness to distribution shifts and adapts effectively to mismatched classifiers.

cs.LG↗

arXiv2025

Low-Regret and Low-Complexity Learning for Hierarchical Inference

This work focuses on Hierarchical Inference (HI) in edge intelligence systems, where a compact Local-ML model on an end-device works in conjunction with a high-accuracy Remote-ML model on an edge-server. HI aims to reduce latency, improve accuracy, and lower bandwidth usage by first using the Local-ML model for inference and offloading to the Remote-ML only when the local inference is likely incorrect. A critical challenge in HI is estimating the likelihood of the local inference being incorrect, especially when data distributions and offloading costs change over time -- a problem we term Hierarchical Inference Learning (HIL). We introduce a novel approach to HIL by modeling the probability of correct inference by the Local-ML as an increasing function of the model's confidence measure, a structure motivated by empirical observations but previously unexploited. We propose two policies, HI-LCB and HI-LCB-lite, based on the Upper Confidence Bound (UCB) framework. We demonstrate that both policies achieve order-optimal regret of $O(\log T)$, a significant improvement over existing HIL policies with $O(T^{2/3})$ regret guarantees. Notably, HI-LCB-lite has an $O(1)$ per-sample computational complexity, making it well-suited for deployment on devices with severe resource limitations. Simulations using real-world datasets confirm that our policies outperform existing state-of-the-art HIL methods.

cs.LG↗

arXiv2025

Error Bounds for the Network Scale-Up Method

Epidemiologists and social scientists have used the Network Scale-Up Method (NSUM) for over thirty years to estimate the size of a hidden sub-population within a social network. This method involves querying a subset of network nodes about the number of their neighbours belonging to the hidden sub-population. In general, NSUM assumes that the social network topology and the hidden sub-population distribution are well-behaved; hence, the NSUM estimate is close to the actual value. However, bounds on NSUM estimation errors have not been analytically proven. This paper provides analytical bounds on the error incurred by the two most popular NSUM estimators. These bounds assume that the queried nodes accurately provide their degree and the number of neighbors belonging to the hidden population. Our key findings are twofold. First, we show that when an adversary designs the network and places the hidden sub-population, then the estimate can be a factor of $Ω(\sqrt{n})$ off from the real value (in a network with $n$ nodes). Second, we also prove error bounds when the underlying network is randomly generated, showing that a small constant factor can be achieved with high probability using samples of logarithmic size $O(\log{n})$. We present improved analytical bounds for Erdos-Renyi and Scale-Free networks. Our theoretical analysis is supported by an extensive set of numerical experiments designed to determine the effect of the sample size on the accuracy of the estimates in both synthetic and real networks.

cs.DC↗

arXiv2025

Towards Efficient Multi-LLM Inference: Characterization and Analysis of LLM Routing and Hierarchical Techniques

Recent progress in Language Models (LMs) has dramatically advanced the field of natural language processing (NLP), excelling at tasks like text generation, summarization, and question answering. However, their inference remains computationally expensive and energy intensive, especially in settings with limited hardware, power, or bandwidth. This makes it difficult to deploy LMs in mobile, edge, or cost sensitive environments. To address these challenges, recent approaches have introduced multi LLM intelligent model selection strategies that dynamically allocate computational resources based on query complexity -- using lightweight models for simpler queries and escalating to larger models only when necessary. This survey explores two complementary strategies for efficient LLM inference: (i) routing, which selects the most suitable model based on the query, and (ii) cascading or hierarchical inference (HI), which escalates queries through a sequence of models until a confident response is found. Both approaches aim to reduce computation by using lightweight models for simpler tasks while offloading only when needed. We provide a comparative analysis of these techniques across key performance metrics, discuss benchmarking efforts, and outline open challenges. Finally, we outline future research directions to enable faster response times, adaptive model selection based on task complexity, and scalable deployment across heterogeneous environments, making LLM based systems more efficient and accessible for real world applications.

cs.LG↗

arXiv2025

Minimizing Age of Detection for a Markov Source over a Lossy Channel

Monitoring a process/phenomenon of specific interest is prevalent in Cyber-Physical Systems (CPS), remote healthcare, smart buildings, intelligent transport, industry 4.0, etc. A key building block of the monitoring system is a sensor sampling the process and communicating the status updates to a monitor for detecting events of interest. Measuring the freshness of the status updates is essential for the timely detection of events, and it has received significant research interest in recent times. In this paper, we propose a new freshness metric, Age of Detection (AoD), for monitoring the state transitions of a Discrete Time Markov Chain (DTMC) source over a lossy wireless channel. We consider the pull model where the sensor samples DTMC state whenever the monitor requests a status update. We formulate a Constrained Markov Decision Problem (CMDP) for optimising the AoD subject to a constraint on the average sampling frequency and solve it using the Lagrangian MDP formulation and Relative Value Iteration (RVI) algorithm. Our numerical results show interesting trade-offs between AoD, sampling frequency, and transmission success probability. Further, the AoD minimizing policy provides a lower estimation error than the Age of Information (AoI) minimizing policy, thus demonstrating the utility of AoD for monitoring DTMC sources.

cs.IT↗

arXiv2024

2D-AoI: Age-of-Information of Distributed Sensors for Spatio-Temporal Processes

The freshness of sensor data is critical for all types of cyber-physical systems. An established measure for quantifying data freshness is the Age-of-Information (AoI), which has been the subject of extensive research. Recently, there has been increased interest in multi-sensor systems: redundant sensors producing samples of the same physical process, sensors such as cameras producing overlapping views, or distributed sensors producing correlated samples. When the information from a particular sensor is outdated, fresh samples from other correlated sensors can be helpful. To quantify the utility of distant but correlated samples, we put forth a two-dimensional (2D) model of AoI that takes into account the sensor distance in an age-equivalent representation. Since we define 2D-AoI as equivalent to AoI, it can be readily linked to existing AoI research, especially on parallel systems. We consider physical phenomena modeled as spatio-temporal processes and derive the 2D-AoI for different Gaussian correlation kernels. For a basic exponential product kernel, we find that spatial distance causes an additive offset of the AoI, while for other kernels the effects of spatial distance are more complex and vary with time. Using our methodology, we evaluate the 2D-AoI of different spatial topologies and sensor densities.

cs.NI↗

arXiv2024

Improved Decision Module Selection for Hierarchical Inference in Resource-Constrained Edge Devices

The Hierarchical Inference (HI) paradigm employs a tiered processing: the inference from simple data samples are accepted at the end device, while complex data samples are offloaded to the central servers. HI has recently emerged as an effective method for balancing inference accuracy, data processing, transmission throughput, and offloading cost. This approach proves particularly efficient in scenarios involving resource-constrained edge devices, such as IoT sensors and micro controller units (MCUs), tasked with executing tinyML inference. Notably, it outperforms strategies such as local inference execution, inference offloading to edge servers or cloud facilities, and split inference (i.e., inference execution distributed between two endpoints). Building upon the HI paradigm, this work explores different techniques aimed at further optimizing inference task execution. We propose and discuss three distinct HI approaches and evaluate their utility for image classification.

cs.DC↗

arXiv2024

Online Algorithms for Hierarchical Inference in Deep Learning applications at the Edge

We consider a resource-constrained Edge Device (ED), such as an IoT sensor or a microcontroller unit, embedded with a small-size ML model (S-ML) for a generic classification application and an Edge Server (ES) that hosts a large-size ML model (L-ML). Since the inference accuracy of S-ML is lower than that of the L-ML, offloading all the data samples to the ES results in high inference accuracy, but it defeats the purpose of embedding S-ML on the ED and deprives the benefits of reduced latency, bandwidth savings, and energy efficiency of doing local inference. In order to get the best out of both worlds, i.e., the benefits of doing inference on the ED and the benefits of doing inference on ES, we explore the idea of Hierarchical Inference (HI), wherein S-ML inference is only accepted when it is correct, otherwise the data sample is offloaded for L-ML inference. However, the ideal implementation of HI is infeasible as the correctness of the S-ML inference is not known to the ED. We propose an online meta-learning framework that the ED can use to predict the correctness of the S-ML inference. In particular, we propose to use the maximum softmax value output by S-ML for a data sample and decide whether to offload it or not. The resulting online learning problem turns out to be a Prediction with Expert Advice (PEA) problem with continuous expert space. We propose two different algorithms and prove sublinear regret bounds for them without any assumption on the smoothness of the loss function. We evaluate and benchmark the performance of the proposed algorithms for image classification application using four datasets, namely, Imagenette and Imagewoof, MNIST, and CIFAR-10.

cs.LG↗

arXiv2023

The Case for Hierarchical Deep Learning Inference at the Network Edge

Resource-constrained Edge Devices (EDs), e.g., IoT sensors and microcontroller units, are expected to make intelligent decisions using Deep Learning (DL) inference at the edge of the network. Toward this end, there is a significant research effort in developing tinyML models - Deep Learning (DL) models with reduced computation and memory storage requirements - that can be embedded on these devices. However, tinyML models have lower inference accuracy. On a different front, DNN partitioning and inference offloading techniques were studied for distributed DL inference between EDs and Edge Servers (ESs). In this paper, we explore Hierarchical Inference (HI), a novel approach proposed by Vishnu et al. 2023, arXiv:2304.00891v1 , for performing distributed DL inference at the edge. Under HI, for each data sample, an ED first uses a local algorithm (e.g., a tinyML model) for inference. Depending on the application, if the inference provided by the local algorithm is incorrect or further assistance is required from large DL models on edge or cloud, only then the ED offloads the data sample. At the outset, HI seems infeasible as the ED, in general, cannot know if the local inference is sufficient or not. Nevertheless, we present the feasibility of implementing HI for machine fault detection and image classification applications. We demonstrate its benefits using quantitative analysis and argue that using HI will result in low latency, bandwidth savings, and energy savings in edge AI systems.

cs.DC↗

arXiv2022

Realistic Modeling of Human Timings for Wearable Cognitive Assistance

Wearable Cognitive Assistance (WCA) applications present a challenge to benchmark and characterize due to their human-in-the-loop nature. Employing user testing to optimize system parameters is generally not feasible, given the scope of the problem and the number of observations needed to detect small but important effects in controlled experiments. Considering the intended mass-scale deployment of WCA applications in the future, there exists a need for tools enabling human-independent benchmarking. We present in this paper the first model for the complete end-to-end emulation of humans in WCA. We build this model through statistical analysis of data collected from previous work in this field, and demonstrate its utility by studying application task durations. Compared to first-order approximations, our model shows a ~36% larger gap between step execution times at high system impairment versus low. We further introduce a novel framework for stochastic optimization of resource consumption-responsiveness tradeoffs in WCA, and show that by combining this framework with our realistic model of human behavior, significant reductions of up to 50% in number processed frame samples and 20% in energy consumption can be achieved with respect to the state-of-the-art.

cs.HC↗

arXiv2022

Edge-MultiAI: Multi-Tenancy of Latency-Sensitive Deep Learning Applications on Edge

Smart IoT-based systems often desire continuous execution of multiple latency-sensitive Deep Learning (DL) applications. The edge servers serve as the cornerstone of such IoT-based systems, however, their resource limitations hamper the continuous execution of multiple (multi-tenant) DL applications. The challenge is that, DL applications function based on bulky "neural network (NN) models" that cannot be simultaneously maintained in the limited memory space of the edge. Accordingly, the main contribution of this research is to overcome the memory contention challenge, thereby, meeting the latency constraints of the DL applications without compromising their inference accuracy. We propose an efficient NN model management framework, called Edge-MultiAI, that ushers the NN models of the DL applications into the edge memory such that the degree of multi-tenancy and the number of warm-starts are maximized. Edge-MultiAI leverages NN model compression techniques, such as model quantization, and dynamically loads NN models for DL applications to stimulate multi-tenancy on the edge server. We also devise a model management heuristic for Edge-MultiAI, called iWS-BFE, that functions based on the Bayesian theory to predict the inference requests for multi-tenant applications, and uses it to choose the appropriate NN models for loading, hence, increasing the number of warm-start inferences. We evaluate the efficacy and robustness of Edge-MultiAI under various configurations. The results reveal that Edge-MultiAI can stimulate the degree of multi-tenancy on the edge by at least 2X and increase the number of warm-starts by around 60% without any major loss on the inference accuracy of the applications.

cs.DC↗

arXiv2021

Offloading Algorithms for Maximizing Inference Accuracy on Edge Device Under a Time Constraint

With the emergence of edge computing, the problem of offloading jobs between an Edge Device (ED) and an Edge Server (ES) received significant attention in the past. Motivated by the fact that an increasing number of applications are using Machine Learning (ML) inference, we study the problem of offloading inference jobs by considering the following novel aspects: 1) in contrast to a typical computational job, the processing time of an inference job depends on the size of the ML model, and 2) recently proposed Deep Neural Networks (DNNs) for resource-constrained devices provide the choice of scaling the model size. We formulate an assignment problem with the aim of maximizing the total inference accuracy of n data samples available at the ED, subject to a time constraint T on the makespan. We propose an approximation algorithm AMR2, and prove that it results in a makespan at most 2T, and achieves a total accuracy that is lower by a small constant from optimal total accuracy. As proof of concept, we implemented AMR2 on a Raspberry Pi, equipped with MobileNet, and is connected to a server equipped with ResNet, and studied the total accuracy and makespan performance of AMR2 for image classification application.

cs.DC↗

arXiv2021

Energy Efficient Sampling Policies for Edge Computing Feedback Systems

We study the problem of finding efficient sampling policies in an edge-based feedback system, where sensor samples are offloaded to a back-end server that processes them and generates feedback to a user. Sampling the system at maximum frequency results in the detection of events of interest with minimum delay but incurs higher energy costs due to the communication and processing of redundant samples. On the other hand, lower sampling frequency results in higher delay in detecting the event, thus increasing the idle energy usage and degrading the quality of experience. We quantify this trade-off as a weighted function between the number of samples and the sampling interval. We solve the minimisation problem for exponential and Rayleigh distributions, for the random time to the event of interest. We prove the convexity of the objective functions by using novel techniques, which can be of independent interest elsewhere. We argue that adding an initial offset to the periodic sampling can further reduce the energy consumption and jointly compute the optimum offset and sampling interval. We apply our framework to two practically relevant applications and show energy savings of up to 36% when compared to an existing periodic scheme.

cs.IT↗

arXiv2021

Estimating Active Cases of COVID-19

Having accurate and timely data on confirmed active COVID-19 cases is challenging, since it depends on testing capacity and the availability of an appropriate infrastructure to perform tests and aggregate their results. In this paper, we propose methods to estimate the number of active cases of COVID-19 from the official data (of confirmed cases and fatalities) and from survey data. We show that the latter is a viable option in countries with reduced testing capacity or suboptimal infrastructures.

physics.soc-ph↗

arXiv2021

Scheduling of Wireless Edge Networks for Feedback-Based Interactive Applications

Interactive applications with automated feedback will largely influence the design of future networked infrastructures. In such applications, status information about an environment of interest is captured and forwarded to a compute node, which analyzes the information and generates a feedback message. Timely processing and forwarding must ensure the feedback information to be still applicable; thus, the quality-of-service parameter for such applications is the end-to-end latency over the entire loop. By modelling the communication of a feedback loop as a two-hop network, we address the problem of allocating network resources in order to minimize the delay violation probability (DVP), i.e. the probability of the end-to-end latency exceeding a target value. We investigate the influence of the network queue states along the network path on the performance of semi-static and dynamic scheduling policies. The former determine the schedule prior to the transmission of the packet, while the latter benefit from feedback on the queue states as time evolves and reallocate time slots depending on the queue's evolution. The performance of the proposed policies is evaluated for variations in several system parameters and comparison baselines. Results show that the proposed semi-static policy achieves close-to-optimal DVP and the dynamic policy outperforms the state-of-the-art algorithms.

cs.NI↗

arXiv2020

Statistical Guarantee Optimization for AoI in Single-Hop and Two-Hop Systems with Periodic Arrivals

Age of Information (AoI) has proven to be a useful metric in networked systems where timely information updates are of importance. In the literature, minimizing "average age" has received considerable attention. However, various applications pose stricter age requirements on the updates which demand knowledge of the AoI distribution. Furthermore, the analysis of AoI distribution in a multi-hop setting, which is important for the study of Wireless Networked Control Systems (WNCS), has not been addressed before. Toward this end, we study the distribution of AoI in a WNCS with two hops and devise a problem of minimizing the tail of the AoI distribution with respect to the frequency of generating information updates, i.e., the sampling rate of monitoring a process, under first-come-first-serve (FCFS) queuing discipline. We argue that computing an exact expression for the AoI distribution may not always be feasible; therefore, we opt for computing upper bounds on the tail of the AoI distribution. Using these upper bounds we formulate Upper Bound Minimization Problems (UBMP), namely, Chernoff-UBMP and alpha-relaxed Upper BoundMinimization Problem (alpha-UBMP), where alpha > 1 is an approximation factor, and solve them to obtain "good" heuristic rate solutions. We demonstrate the efficacy of our approach by solving the proposed UBMPs for three service distributions: geometric, exponential, and Erlang. Simulation results show that the rate solutions obtained are near-optimal for minimizing the tail of the AoI distribution for the considered distributions.

cs.IT↗