arxiv: 2604.16448 · v1 · submitted 2026-04-07 · 📡 eess.SY · cs.LG· cs.SY

Recognition: 2 theorem links

· Lean Theorem

FM-CAC: Carbon-Aware Control for Battery-Buffered Edge AI via Time-Series Foundation Models

Kang Yang , Walid A. Hanafy , Prashant Shenoy , Mani Srivastava

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:17 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SY

keywords fm-caccarbonemissionsenergybatterycontrolcarbon-awareconsumption

0 comments

The pith

FM-CAC uses battery buffering and time-series foundation models for zero-shot carbon forecasting in a dynamic programming optimizer to reduce edge AI carbon emissions by up to 65.6% with near-maximum accuracy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Edge AI runs on many small devices that stay on all the time, using electricity that often produces carbon emissions. FM-CAC adds a battery to each device so it can store clean energy when available and use it later when grid power is dirtier. To decide the best times, it uses large time-series foundation models that predict future carbon levels without needing location-specific training data. These predictions feed into a dynamic programming planner that also picks the right AI model version and hardware speed. A technique called deferred cost attribution stops the planner from draining the battery too fast in the short term. The system treats energy use, battery charge, and AI settings as one joint decision at every step. The abstract reports that this approach cut carbon output by as much as 65.6 percent while keeping inference accuracy almost as high as the best possible setting.

Core claim

Results show that FM-CAC reduces carbon emissions by up to 65.6% while maintaining near-maximum inference accuracy.

Load-bearing premise

That time-series foundation models deliver sufficiently accurate zero-shot carbon forecasts and that battery charge/discharge actions can be optimized jointly with pipeline and hardware choices without violating QoS constraints.

Figures

Figures reproduced from arXiv: 2604.16448 by Kang Yang, Mani Srivastava, Prashant Shenoy, Walid A. Hanafy.

**Figure 2.** Figure 2: Comparison of FM-CAC (Ours) and four baselines across four key metrics in the CAISO region over the test period. ranges from 4.1 W to 10.0 W. This profiling defines a joint mode space O = N × H with 600 distinct model–hardware pairs. We further impose strict QoS constraints, requiring a minimum accuracy of 𝑢acc = 0.40 mAP50–95 and a maximum latency of 𝑢lat = 100 ms. Battery. We model a compact consumer-gr… view at source ↗

**Figure 4.** Figure 4: Impact of region. 5 10 18 20 30 40 50 Different Battery Capacity (Wh) 0 25 50 75 100 C a r b o n ( g C O 2 ) 0.51 0.52 0.53 0.54 0.54 Accuracy (mAP50-95) [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

read the original abstract

As edge AI deployments scale to billions of devices running always-on, real-time compound AI pipelines, they represent a massive and largely unmanaged source of energy consumption and carbon emissions. To reduce carbon emissions while maximizing Quality-of-Service (QoS), this paper proposes FM-CAC, a proactive carbon-aware control framework that leverages a battery as an active temporal buffer. By decoupling energy acquisition from energy consumption, FM-CAC can maximize the use of low-carbon energy, substantially reducing carbon emissions. At each control step, FM-CAC jointly optimizes the software pipeline variant, the hardware operating point, and the battery charging and discharging actions. To support this decision process, FM-CAC leverages edge-friendly Time-Series Foundation Models (TSFMs) for zero-shot carbon forecasting and integrates these forecasts into a dynamic programming solver with deferred cost attribution to prevent myopic battery depletion. Results show that FM-CAC reduces carbon emissions by up to 65.6% while maintaining near-maximum inference accuracy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

FM-CAC folds zero-shot TSFM carbon forecasts into a deferred-cost DP that jointly picks battery actions, pipeline variants, and hardware points for edge AI, but the 65.6% savings figure needs forecast-error checks to be convincing.

read the letter

The paper presents FM-CAC as a way to make edge AI more carbon-aware by treating the battery as a buffer that lets the system wait for low-carbon energy periods. It jointly decides on the AI pipeline variant, hardware settings, and when to charge or discharge the battery. The forecasts come from time-series foundation models used in zero-shot mode, and those feed into a dynamic program that uses deferred cost attribution to avoid myopic decisions. What stands out is the specific integration. Prior work has looked at carbon-aware scheduling or battery use separately, but putting TSFM forecasts directly into this joint optimization with the deferred cost trick seems fresh. The approach is practical for always-on compound pipelines on battery-buffered devices. The results claim up to 65.6% emission reduction while keeping accuracy high. That would be a strong outcome if it generalizes. The framework itself is clearly described and seems logically consistent. The soft spot is around the forecast quality and its impact. Without reported forecast errors, ablations against baselines like persistence, or sensitivity on how errors affect the DP outcomes, the big savings number is hard to trust fully. If the TSFM is off on the carbon traces used for evaluation, the realized savings could be lower or QoS could suffer. I would want to see those details in the full paper to judge how solid the claim is. This paper is aimed at researchers in sustainable edge computing and systems for AI. Someone building real deployments or studying carbon optimization would find the ideas useful to build on or test further. It deserves peer review. The problem is important and the proposed synthesis is worth a closer look from referees who can examine the experiments and any additional validation provided.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes FM-CAC, a proactive carbon-aware control framework for battery-buffered edge AI systems. It decouples energy acquisition from consumption via a battery buffer and jointly optimizes software pipeline variants, hardware operating points, and battery charge/discharge actions at each control step. Zero-shot carbon-intensity forecasts from edge-friendly time-series foundation models (TSFMs) are integrated into a dynamic programming solver that uses deferred cost attribution to avoid myopic battery depletion. The central empirical claim is that FM-CAC achieves up to 65.6% carbon-emission reduction while preserving near-maximum inference accuracy.

Significance. If the experimental results are reproducible and the forecast accuracy is shown to be sufficient, the work would be significant for sustainable edge computing. It demonstrates a concrete control architecture that combines foundation-model forecasting with optimal control under QoS constraints, addressing a timely problem as edge AI deployments scale. The use of TSFMs for zero-shot forecasting and the deferred-cost DP formulation are technically interesting contributions that could influence future carbon-aware system designs.

major comments (3)

[Abstract, §5] Abstract and §5 (Results): The headline claim of a 65.6% carbon reduction is stated without any reported baselines, statistical tests, error bars, or ablation studies. No forecast MAE, persistence/oracle comparisons, or sensitivity analysis showing how TSFM error propagates into realized savings and QoS violations is provided, leaving the central empirical result unsupported.
[§3.2, §4] §3.2 (Forecast Integration) and §4 (Optimization): The manuscript asserts that TSFM zero-shot forecasts are sufficiently accurate to drive the DP solver, yet supplies no quantitative validation of forecast quality on the evaluation traces nor any analysis of how forecast error affects battery scheduling decisions or QoS constraint satisfaction.
[§4] §4 (Dynamic Programming): The deferred-cost attribution mechanism is introduced to prevent myopic depletion, but the paper does not demonstrate that the resulting policy remains feasible under realistic forecast errors or that the claimed savings are robust to the joint optimization of pipeline, hardware, and battery actions.

minor comments (2)

[§2] Notation for carbon intensity, battery state, and QoS metrics should be defined consistently in a single table or early section to improve readability.
[Abstract, §1] The abstract and introduction would benefit from a brief statement of the evaluation traces, hardware platforms, and carbon-intensity datasets used.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the positive assessment of FM-CAC's significance for sustainable edge computing and for the constructive comments on empirical validation. We address each major comment below and will revise the manuscript to incorporate the requested analyses and demonstrations.

read point-by-point responses

Referee: [Abstract, §5] Abstract and §5 (Results): The headline claim of a 65.6% carbon reduction is stated without any reported baselines, statistical tests, error bars, or ablation studies. No forecast MAE, persistence/oracle comparisons, or sensitivity analysis showing how TSFM error propagates into realized savings and QoS violations is provided, leaving the central empirical result unsupported.

Authors: We agree that the 65.6% reduction figure requires additional context to be fully supported. In the revised manuscript we will expand §5 with comparisons to baselines including no-battery operation, greedy scheduling, persistence forecasting, and an oracle with perfect information. Results will include error bars from repeated runs across traces, statistical significance tests (e.g., paired t-tests), TSFM forecast MAE on the evaluation carbon traces, and a sensitivity analysis that perturbs forecasts to quantify effects on realized savings and QoS violations. These additions will directly substantiate the central claim. revision: yes
Referee: [§3.2, §4] §3.2 (Forecast Integration) and §4 (Optimization): The manuscript asserts that TSFM zero-shot forecasts are sufficiently accurate to drive the DP solver, yet supplies no quantitative validation of forecast quality on the evaluation traces nor any analysis of how forecast error affects battery scheduling decisions or QoS constraint satisfaction.

Authors: We acknowledge the absence of explicit forecast validation in the current text. The revision will add quantitative metrics (MAE and RMSE) for the TSFM zero-shot forecasts on the specific carbon-intensity traces used in the experiments. We will also include a direct comparison of DP-derived battery schedules and QoS outcomes under TSFM forecasts versus oracle and persistence forecasts, thereby quantifying how forecast error influences scheduling decisions and constraint satisfaction. revision: yes
Referee: [§4] §4 (Dynamic Programming): The deferred-cost attribution mechanism is introduced to prevent myopic depletion, but the paper does not demonstrate that the resulting policy remains feasible under realistic forecast errors or that the claimed savings are robust to the joint optimization of pipeline, hardware, and battery actions.

Authors: We appreciate the call for explicit robustness checks on the deferred-cost DP. In the revised evaluation we will run the solver under forecast perturbations matching observed TSFM error levels and report battery-state feasibility together with QoS satisfaction rates. We will further add ablation studies that isolate the joint optimization over pipeline variants, hardware points, and battery actions, demonstrating incremental benefits and confirming that the claimed savings remain stable under the full optimization. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical result from external TSFM + DP solver

full rationale

The paper's derivation chain consists of feeding zero-shot forecasts from an external time-series foundation model into a standard dynamic programming solver that jointly selects pipeline variants, hardware points, and battery actions under deferred cost attribution. The headline 65.6% emission reduction is presented as an observed outcome on evaluation traces rather than a quantity obtained by algebraic rearrangement or by fitting a parameter to the same data it is later claimed to predict. No self-definitional equations, no fitted inputs relabeled as predictions, and no load-bearing self-citations appear in the described framework. The result therefore remains dependent on the independent accuracy of the TSFM forecasts and the solver's optimization behavior, both of which are external to the paper's own definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract provides no explicit free parameters, axioms, or invented entities; the framework is described as building on existing TSFMs and dynamic programming without introducing new postulates or fitted constants.

pith-pipeline@v0.9.0 · 5488 in / 1152 out tokens · 54976 ms · 2026-05-10T19:17:00.483378+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

FM-CAC jointly optimizes the software pipeline variant, the hardware operating point, and the battery charging and discharging actions... integrates these forecasts into a dynamic programming solver with deferred cost attribution
IndisputableMonolith/Foundation/ArrowOfTime.lean arrow_from_z unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

deferred cost attribution mechanism that internalizes the anticipated carbon cost of future recharge

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

57 extracted references · 7 canonical work pages · 2 internal anchors

[1]

Carbon-Aware Computing for Datacenters.IEEE Transactions on Power Systems, 38(2):1270–1280, 2022

Ana Radovanović, Ross Koningstein, Ian Schneider, Bokan Chen, Alexandre Duarte, Binz Roy, Diyue Xiao, Maya Haridasan, Patrick Hung, Nick Care, et al. Carbon-Aware Computing for Datacenters.IEEE Transactions on Power Systems, 38(2):1270–1280, 2022

2022
[2]

State of IoT 2025: Number of Connected IoT Devices Growing 14% to 21.1 Billion Globally

IoT Analytics. State of IoT 2025: Number of Connected IoT Devices Growing 14% to 21.1 Billion Globally. https://iot-analytics.com/number-connected-iot- devices/, 2025. [Online]

2025
[3]

How Many Smartphones Are in the World? https://www

BankMyCell. How Many Smartphones Are in the World? https://www. bankmycell.com/blog/how-many-phones-are-in-the-world, 2026. [Online]

2026
[4]

Study: One Billion Surveillance Cameras Expected to Be In- stalled Globally by 2021

Haley Samsel. Study: One Billion Surveillance Cameras Expected to Be In- stalled Globally by 2021. https://securitytoday.com/articles/2019/12/09/one- billion-surveillance-cameras-expected.aspx, 2019. [Online]

2021
[5]

Environmental Protection Agency (EPA)

U.S. Environmental Protection Agency (EPA). Greenhouse Gas Emissions from a Typical Passenger Vehicle. https://www.epa.gov/greenvehicles/greenhouse-gas- emissions-typical-passenger-vehicle, 2023. [Online]

2023
[6]

Statues: Energy-Efficient Video Object Detection on Edge Security Devices with Computational Skipping

Yeonggeon Kim, Hyunmin Kim, and Sungju Ryu. Statues: Energy-Efficient Video Object Detection on Edge Security Devices with Computational Skipping. In ACM/IEEE ISLPED, 2024

2024
[7]

Nest Cam with Floodlight Technical Specs

Google. Nest Cam with Floodlight Technical Specs. https://store.google.com/ product/nest_cam_floodlight_specs, 2024. [Online]

2024
[8]

The Shift from Models to Compound AI Systems

Matei Zaharia et al. The Shift from Models to Compound AI Systems. https: //bair.berkeley.edu/blog/2024/02/18/compound-ai-systems/, 2024. [Online]

2024
[9]

Navigating Battery Choices in IoT: An Extensive Survey of Technologies and Their Applications.Batteries, 9(12), 2023

Kareeb Hasan, Neil Tom, and Mehmet Rasit Yuce. Navigating Battery Choices in IoT: An Extensive Survey of Technologies and Their Applications.Batteries, 9(12), 2023

2023
[10]

KONA Micro IoT Gateway

TEKTELIC Communications. KONA Micro IoT Gateway. https://tektelic.com/ products/gateways/kona-micro-iot-gateway/, 2024. [Online]

2024
[11]

Energy- Aware Dynamic Neural Inference.arXiv preprint arXiv:2411.02471, 2024

Marcello Bullo, Seifallah Jardak, Pietro Carnelli, and Deniz Gündüz. Energy- Aware Dynamic Neural Inference.arXiv preprint arXiv:2411.02471, 2024

work page arXiv 2024
[12]

E4: Energy-Efficient DNN Inference for Edge Video Analytics via Early Exiting and DVFS

Ziyang Zhang, Yang Zhao, Ming-Ching Chang, Changyao Lin, and Jie Liu. E4: Energy-Efficient DNN Inference for Edge Video Analytics via Early Exiting and DVFS. InAAAI, 2025

2025
[13]

HarvNet: Resource-Optimized Operation of Multi-Exit Deep Neural Networks on Energy Harvesting Devices

Seunghyeok Jeon, Yonghun Choi, Yeonwoo Cho, and Hojung Cha. HarvNet: Resource-Optimized Operation of Multi-Exit Deep Neural Networks on Energy Harvesting Devices. InACM MobiSys, 2023

2023
[14]

DVFS-Aware DNN Inference on GPUs: Latency Modeling and Performance Analysis

Yunchu Han, Zhaojun Nan, Sheng Zhou, and Zhisheng Niu. DVFS-Aware DNN Inference on GPUs: Latency Modeling and Performance Analysis. InIEEE ICC, 2025

2025
[15]

Energy-Efficient Inference with Software–Hardware Co-Design for Sustainable Artificial Intelligence of Things.IEEE Internet of Things Journal, 11(24):39170–39182, 2024

Shengxin Dai, Zheng Luo, Wendian Luo, Siyi Wang, Cheng Dai, Bing Guo, and Xiaokang Zhou. Energy-Efficient Inference with Software–Hardware Co-Design for Sustainable Artificial Intelligence of Things.IEEE Internet of Things Journal, 11(24):39170–39182, 2024

2024
[16]

Energy Efficient and QoS-Aware Model Selection for DNN Inference in Edge Intelligence

Hajar Siar and Erik Elmroth. Energy Efficient and QoS-Aware Model Selection for DNN Inference in Edge Intelligence. InIEEE CloudCom, 2025

2025
[17]

Greening Edge AI: Optimizing Inference Accuracy and Reducing Carbon Emissions with Renewable Energy.IEEE Internet of Things Journal, 2025

Huirong Ma, Zhilan Huang, Gang Lu, Deji Fu, and Chunyu Shi. Greening Edge AI: Optimizing Inference Accuracy and Reducing Carbon Emissions with Renewable Energy.IEEE Internet of Things Journal, 2025

2025
[18]

Does Energy-Efficiency Mean Carbon-Awareness? A Critical Evaluation of Sustainable Serverless Edge Scheduling

Shaoshu Zhu, Neil Hurley, and Hadi Tabatabaee Malazi. Does Energy-Efficiency Mean Carbon-Awareness? A Critical Evaluation of Sustainable Serverless Edge Scheduling. InIEEE/ACM UCC, 2025

2025
[19]

Adaptive Duty Cycling for Energy Harvesting Systems

Jason Hsu, Sadaf Zahedi, Aman Kansal, Mani Srivastava, and Vijay Raghunathan. Adaptive Duty Cycling for Energy Harvesting Systems. InACM/IEEE ISLPED, 2006

2006
[20]

CLOVER: Carbon Optimization of Federated Learning over Heterogeneous Clients

Chanwoo Cho, Yonglak Son, Seongbin Park, and Young Geun Kim. CLOVER: Carbon Optimization of Federated Learning over Heterogeneous Clients. In ACM/IEEE ISLPED, 2024

2024
[21]

DiReCt: Resource-Aware Dynamic Model Reconfiguration for Convolutional Neural Net- work in Mobile Systems

Zirui Xu, Zhuwei Qin, Fuxun Yu, Chenchen Liu, and Xiang Chen. DiReCt: Resource-Aware Dynamic Model Reconfiguration for Convolutional Neural Net- work in Mobile Systems. InACM/IEEE ISLPED, 2018

2018
[22]

Hanafy, Noman Bashir, Ahmed Ali-Eldin, David Irwin, and Prashant Shenoy

Qianlin Liang, Walid A. Hanafy, Noman Bashir, Ahmed Ali-Eldin, David Irwin, and Prashant Shenoy. DěLen: Enabling Flexible and Adaptive Model-Serving for Multi-Tenant Edge AI. InACM/IEEE IoTDI, 2023

2023
[23]

Energy-Harvesting-Aware Adaptive Inference of Deep Neural Networks in Embedded Systems

Gwanjong Park, Osama Khan, and Euiseong Seo. Energy-Harvesting-Aware Adaptive Inference of Deep Neural Networks in Embedded Systems. InACM/IEEE ISLPED, 2023

2023
[24]

Let’s Wait Awhile: How Temporal Workload Shifting Can Reduce Carbon Emissions in the Cloud

Philipp Wiesner, Ilja Behnke, Dominik Scheinert, Kordian Gontarska, and Lauritz Thamsen. Let’s Wait Awhile: How Temporal Workload Shifting Can Reduce Carbon Emissions in the Cloud. InACM Middleware, 2021

2021
[25]

Bring- ing Carbon Awareness to Multi-Cloud Application Delivery

Diptyaroop Maji, Ben Pfaff, Vipin PR, Rajagopal Sreenivasan, Victor Firoiu, Sreeram Iyer, Colleen Josephson, Zhelong Pan, and Ramesh K Sitaraman. Bring- ing Carbon Awareness to Multi-Cloud Application Delivery. InHotCarbon, 2023. Conference acronym ’XX, June 03–05, 2026, Woodstock, NY Kang Yang, Walid A. Hanafy, Prashant Shenoy, and Mani Srivastava

2023
[26]

EcoLife: Carbon- Aware Serverless Function Scheduling for Sustainable Computing

Yankai Jiang, Rohan Basu Roy, Baolin Li, and Devesh Tiwari. EcoLife: Carbon- Aware Serverless Function Scheduling for Sustainable Computing. InACM/IEEE SC, 2024

2024
[27]

GreenScale: Carbon-Aware Systems for Edge Computing.arXiv preprint arXiv:2304.00404, 2023

Young Geun Kim, Udit Gupta, Andrew McCrabb, Yonglak Son, Valeria Bertacco, David Brooks, and Carole-Jean Wu. GreenScale: Carbon-Aware Systems for Edge Computing.arXiv preprint arXiv:2304.00404, 2023

work page arXiv 2023
[28]

Learning to Be Green: Carbon-Aware Online Control for Edge Intelligence with Colocated Learning and Inference

Shuomiao Su, Zhi Zhou, Tao Ouyang, Ruiting Zhou, and Xu Chen. Learning to Be Green: Carbon-Aware Online Control for Edge Intelligence with Colocated Learning and Inference. InIEEE ICDCS, 2023

2023
[29]

Online Smoothed Demand Management.arXiv preprint arXiv:2511.18554, 2025

Adam Lechowicz, Nicolas Christianson, Mohammad Hajiesmaili, Adam Wierman, and Prashant Shenoy. Online Smoothed Demand Management.arXiv preprint arXiv:2511.18554, 2025

work page arXiv 2025
[30]

Ahmad Faruqui and Sanem Sergici. Will Smart Prices Induce Smart Charging of Electric Vehicles? https://www.brattle.com/wp-content/uploads/2017/10/6368_ will_smart_prices_induce_smart_charging_of_electric_vehicles_jul_2011.pdf,

2017
[31]

Impact of Electric Vehicle Customer Response to Time-Of-Use Rates on Distribution Power Grids.Energy Reports, 8:8225–8235, 2022

C Birk Jones, William Vining, Matthew Lave, Thad Haines, Christopher Neuman, Jesse Bennett, and Don R Scoffield. Impact of Electric Vehicle Customer Response to Time-Of-Use Rates on Distribution Power Grids.Energy Reports, 8:8225–8235, 2022

2022
[32]

Carbon-Aware EV Charging

Kai-Wen Cheng, Yuexin Bian, Yuanyuan Shi, and Yize Chen. Carbon-Aware EV Charging. InIEEE SmartGridComm, 2022

2022
[33]

Scheduling Electric Vehicle Charging to Minimise Carbon Emissions and Wind Curtailment

James Dixon, Waqquas Bukhsh, Calum Edmunds, and Keith Bell. Scheduling Electric Vehicle Charging to Minimise Carbon Emissions and Wind Curtailment. Renewable Energy, 161:1072–1091, 2020

2020
[34]

Smart Charging Strategy for Electric Vehicles Based on Marginal Carbon Emission Factors and Time-Of- Use Price.Sustainable Cities and Society, 96:104708, 2023

Jiale Li, Guilin Wang, Xuefei Wang, and Yingkang Du. Smart Charging Strategy for Electric Vehicles Based on Marginal Carbon Emission Factors and Time-Of- Use Price.Sustainable Cities and Society, 96:104708, 2023

2023
[35]

MOMENT: A Family of Open Time-Series Foundation Models

Mononito Goswami, Konrad Szafer, Arjun Choudhry, Yifu Cai, Shuo Li, and Artur Dubrawski. MOMENT: A Family of Open Time-Series Foundation Models. In ICML, 2024

2024
[36]

Chronos: Learning the Language of Time Series

Abdul Fatir Ansari et al. Chronos: Learning the Language of Time Series.arXiv preprint arXiv:2403.07815, 2024

work page internal anchor Pith review arXiv 2024
[37]

Sitaraman, and Mani Srivastava

Diptyaroop Maji, Kang Yang, Prashant Shenoy, Ramesh K. Sitaraman, and Mani Srivastava. CarbonX: An Open-Source Tool for Computational Decarbonization Using Time Series Foundation Models. InACM e-Energy, 2026

2026
[38]

Benchmarking spatiotemporal reasoning in llms and reasoning models: Capabilities and challenges.arXiv preprint arXiv:2505.11618, 2025

Pengrui Quan, Brian Wang, Kang Yang, Liying Han, and Mani Srivastava. Bench- marking Spatiotemporal Reasoning in LLMs and Reasoning Models: Capabilities and Challenges.arXiv preprint arXiv:2505.11618, 2025

work page arXiv 2025
[39]

Sundial: A Family of Highly Capable Time Series Foundation Models

Yong Liu, Guo Qin, Zhiyuan Shi, Zhi Chen, Caiyin Yang, Xiangdong Huang, Jianmin Wang, and Mingsheng Long. Sundial: A Family of Highly Capable Time Series Foundation Models.arXiv preprint arXiv:2502.00816, 2025

work page internal anchor Pith review arXiv 2025
[40]

Jetson – Embedded AI Computing Platform

NVIDIA. Jetson – Embedded AI Computing Platform. https://developer.nvidia. com/embedded-computing, 2026. Accessed: 2026-03-16

2026
[41]

Neural Network Accelerator with Parameters Resident on Chip, December 10

Olivier Temam, Harshit Khaitan, Ravi Narayanaswami, and Dong Hyuk Woo. Neural Network Accelerator with Parameters Resident on Chip, December 10
[42]

Patent# US20190050717A1
[43]

Deep Learning on Mobile Devices through Neural Processing Units and Edge Computing

Tianxiang Tan and Guohong Cao. Deep Learning on Mobile Devices through Neural Processing Units and Edge Computing. InIEEE INFOCOM, 2022

2022
[44]

Does Accurate Real-Time AI Need Edge Offload? InSEC, 2025

Qifei Dong, Jingao Xu, Padmanabhan Pillai, and Mahadev Satyanarayanan. Does Accurate Real-Time AI Need Edge Offload? InSEC, 2025

2025
[45]

Battery Pack Prices Drop

Jake Hertz. Battery Pack Prices Drop. https://www.batterytechonline.com/ trends/battery-pack-prices-drop-8-to-record-108kwh-despite-rising-lithium- and-cobalt-costs-in-2025, 2025. [Online]

2025
[46]

Accuracy-Resource Tradeoff for Edge Devices in Internet of Things

Nima Mousavi, Baris Aksanli, Alper Sinan Akyurek, and Tajana Simunic Rosing. Accuracy-Resource Tradeoff for Edge Devices in Internet of Things. InIEEE PerCom Workshops, 2017

2017
[47]

Carbon Footprint Distributions of Lithium-Ion Batteries and Their Materials.Nature Communications, 15(1):10301, 2024

Leopold Peiseler, Vanessa Schenker, Karin Schatzmann, Stephan Pfister, Vanessa Wood, and Tobias Schmidt. Carbon Footprint Distributions of Lithium-Ion Batteries and Their Materials.Nature Communications, 15(1):10301, 2024

2024
[48]

A Greener Edge: A Framework on Carbon-Aware Edge ML System Design

Xuesi Chen, Ilan Mandel, Eren Yildiz, Josiah Hester, and Udit Gupta. A Greener Edge: A Framework on Carbon-Aware Edge ML System Design. InACM MobiSys, 2026

2026
[49]

Hanafy, Abel Souza, Khai Nguyen, Jan Harkes, David Irwin, Ma- hadev Satyanarayanan, and Prashant Shenoy

Li Wu, Walid A. Hanafy, Abel Souza, Khai Nguyen, Jan Harkes, David Irwin, Ma- hadev Satyanarayanan, and Prashant Shenoy. CarbonEdge: Leveraging Mesoscale Spatial Carbon-Intensity Variations for Low Carbon Edge Computing. InHPDC, 2025

2025
[50]

Hanafy, David Irwin, and Prashant Shenoy

Phuthipong Bovornkeeratiroj, Walid A. Hanafy, David Irwin, and Prashant Shenoy. GreenThrift: Optimizing Carbon and Cost for Flexible Residential Loads. ACM Journal on Computing and Sustainable Societies, 2025

2025
[51]

Carbonscaler: Leveraging Cloud Workload Elasticity for Optimizing Carbon- Efficiency.Proceedings of the ACM on Measurement and Analysis of Computing Systems, 7(3):1–28, 2023

Walid A Hanafy, Qianlin Liang, Noman Bashir, David Irwin, and Prashant Shenoy. Carbonscaler: Leveraging Cloud Workload Elasticity for Optimizing Carbon- Efficiency.Proceedings of the ACM on Measurement and Analysis of Computing Systems, 7(3):1–28, 2023

2023
[52]

Microsoft COCO: Common Objects in Context

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft COCO: Common Objects in Context. InECCV, 2014

2014
[53]

Battery Management System: An Overview of Its Application in the Smart Grid and Electric Vehicles.IEEE Industrial Electronics Magazine, 7(2):4–16, 2013

Habiballah Rahimi-Eichi, Unnati Ojha, Federico Baronti, and Mo-Yuen Chow. Battery Management System: An Overview of Its Application in the Smart Grid and Electric Vehicles.IEEE Industrial Electronics Magazine, 7(2):4–16, 2013

2013
[54]

Peukert’s Law of a Lead-Acid Battery Simulated by a Mathematical Model.Electrochemical Society Transactions, 25(35):223–233, 2010

Mikäel G Cugnet, Matthieu Dubarry, and Bor Yann Liaw. Peukert’s Law of a Lead-Acid Battery Simulated by a Mathematical Model.Electrochemical Society Transactions, 25(35):223–233, 2010

2010
[55]

Spectral predictability as a fast reliability indicator for time series forecasting model selection,

Oliver Wang, Pengrui Quan, Kang Yang, and Mani Srivastava. Spectral Predictabil- ity as a Fast Reliability Indicator for Time Series Forecasting Model Selection. arXiv preprint arXiv:2511.08884, 2025

work page arXiv 2025
[56]

iPhone Batteries Compared: Capacity, mAh and Watt Hours for Every iPhone

Simon Jary. iPhone Batteries Compared: Capacity, mAh and Watt Hours for Every iPhone. https://www.macworld.com/article/678413/iphone-battery-capacities- compared-all-iphones-battery-life-in-mah-and-wh.html, 2026. [Online]

2026
[57]

Bench- marking Deep Learning Models for Object Detection on Edge Computing Devices

Daghash K Alqahtani, Muhammad Aamir Cheema, and Adel N Toosi. Bench- marking Deep Learning Models for Object Detection on Edge Computing Devices. InICSOC, 2024

2024