Profiling AI Models for Efficient Computation Offloading in Heterogeneous Edge AI Systems

Table of Contents

• 1. Gabatarwa
• 2. Taswirar Bincike
  • 2.1 Local AI Model Profiling
  • 2.2 Resource and Time Prediction
  • 2.3 Task Offloading and Scheduling
• 3. Technical Details
  • 3.1 Mathematical Formulations
  • 3.2 Code Implementation
• 4. Sakamakon Gwaji
• 5. Original Analysis
• 6. Future Applications and Directions
• 7. Nassoshi

1. Gabatarwa

Haɓakar saurin aikace-aikacen AI na mai ƙarewa, kamar fahimtar hoto na lokaci-lokaci da AI na samarwa, ya haifar da manyan buƙatun bayanai da sarrafawa waɗanda galibi suka wuce iyawar na'urar. Edge AI yana magance waɗannan ƙalubalen ta hanyar jujjuya lissafi zuwa gefen cibiyar sadarwa, inda sarrafa AI mai saurin na'ura zai iya faruwa. Wannan hanya tana da mahimmanci ga AI da RAN, wani muhimmin sashi na cibiyoyin sadarwar 6G na gaba kamar yadda Ƙungiyar AI-RAN ta zayyana. A cikin 6G, haɗin AI a ko'ina cikin gefe-RAN da na'urori masu tsattsauran ra'ayi zai tallafa wa rarraba bayanai mai inganci da dabarun AI da aka rarraba, haɓaka sirri da rage jinkiri don aikace-aikace kamar Metaverse da tiyata mai nisa.

Duk da waɗannan fa'idodin, Edge AI yana fuskantar ƙalubale. Ƙarancin wadatar albarkatun a gefe na iya kawo cikas ga aiki yayin jujjuyawar lokaci guda. Bugu da ƙari, zato na tsarin tsarin gine-gine iri ɗaya a cikin wallafe-wallafen da ke akwai ba gaskiya bane, saboda na'urorin gefe sun bambanta sosai a cikin saurin sarrafawa da gine-gine (misali, 1.5GHz da 3.5GHz, ko X86 da ARM), yana tasiri sarrafa aiki da amfani da albarkatun.

2. Taswirar Bincike

Taswirar binciken mu ta mayar da hankali ne kan bayanin samfuran AI don inganta aikin cirewa a cikin tsarin AI na gefe daban-daban. Tsarin ya ƙunshi saitin tsarin, bayanin samfurin AI, horar da samfurin rarraba, manufofin cirewa, da tsarin aiki.

2.1 Local AI Model Profiling

Wannan mataki yana nazarin yadda ƙwayoyin na'ura da halayen tsarin ke tasiri aikin samfurin AI a cikin saitunan kayan aiki daban-daban. Manufar ita ce gano alaƙa tsakanin abubuwa kamar nau'ikan samfurin AI (MLP, CNN), ma'auni (saurin koyo, mai ingantawa), ƙayyadaddun kayan aiki (tsarin gine-gine, FLOPS), da halayen bayanai (girman, girman rukuni), da tasirinsu akan daidacin samfurin, amfani da albarkatu, da lokacin kammala aiki.

2.2 Resource and Time Prediction

Ta amfani da bayanan bayanan aiki, muna hasashen bukatun albarkatu da lokutan kammala ayyuka don ba da damar shirya ayyuka cikin inganci a cikin nodes na gefe. Ana amfani da dabaru kamar XGBoost don cimma ingantaccen hasashe.

2.3 Task Offloading and Scheduling

Bisa ga ra'ayoyin, ana jefa ayyuka a waje kuma ana tsara su don inganta rabon albarkatu da haɓaka aikin Edge AI a cikin yanayi daban-daban.

3. Technical Details

3.1 Mathematical Formulations

Manyan dabaru sun haɗa da daidaitaccen RMSE don daidaiton hasashe: $NRMSE = \frac{\sqrt{\frac{1}{n} \sum_{i=1}^{n} (y_i - \hat{y}_i)^2}}{y_{\max} - y_{\min}}$, inda $y_i$ shine ainilin ƙima, $\hat{y}_i$ shine ƙimar da aka annabta, kuma $y_{\max} - y_{\min}$ shine kewayon ainilai. Ana amfani da amfani da albarkatun azaman $R = f(M, H, D)$, inda $M$ shine nau'in samfuri, $H$ shine ƙayyadaddun kayan aiki, kuma $D$ shine sifofin bayanai.

3.2 Code Implementation

Pseudocode don tsarin aikin tantancewa:

def ai_model_profiling(model_type, hyperparams, hardware_specs, dataset):

4. Sakamakon Gwaji

An farko gwajin ya ƙunshi gudun sama da 3,000 tare da sassa daban-daban. Ta amfani da XGBoost don hasashe, mun sami RMSE mai daidaitaccen ƙima na 0.001, wani gagarumin ci gaba akan MLPs masu ɗauke da fiye da miliyan 4 sigogi. Wannan yana nuna ingancin hanyarmu ta yin bayanin martaba wajen inganta rabon albarkatu da haɓaka aikin Edge AI.

Hoton 1 yana kwatanta taswirar hanyar bincike, yana nuna kwararar daga saitin na'urar zuwa tsara ayyuka, yana nuna haɗa bayanan martaba cikin manufofin juyar da kaya.

5. Original Analysis

Wannan bincike yana gabatar da ci gaba mai mahimmanci a cikin Edge AI ta hanyar magance bambance-bambancen na'urorin gefe ta hanyar yin la'akari da tsarin samfuran AI. Hanyar ta yi daidai da hangen nesa na AI-RAN Alliance na hanyoyin sadarwa na 6G, inda ingantaccen juzu'in lissafi ke da mahimmanci ga aikace-aikacen masu kula da jinkiri kamar motocin cin gashin kansu da haɓaka gaskiya. Amfani da XGBoost don hasashen albarkatun, samun daidaitaccen RMSE na 0.001, ya fi hanyoyin gargajiya kamar MLPs, kama da haɓakar da ake gani a cikin CycleGAN don ayyukan fassarar hoto (Zhu et al., 2017). Wannan ingancin yana da mahimmanci ga tsarin ainihin-lokaci inda ƙuntatawa albarkatu suke da mahimmanci, kamar yadda aka lura a cikin bincike daga IEEE Edge Computing Consortium.

The profiling methodology captures dependencies between model hyperparameters, hardware specs, and performance metrics, enabling predictive scheduling. This is akin to reinforcement learning techniques in distributed systems, such as those explored by Google Research for data center optimization. However, the focus on bare-metal edge environments adds a layer of complexity due to hardware variability, which is often overlooked in homogeneous cloud-based AI systems. The integration with 6G infrastructure promises enhanced privacy and reduced latency, supporting emerging applications like the Metaverse. Future work could explore federated learning integration, as proposed by Konečný et al. (2016), to further improve data privacy while maintaining profiling accuracy.

Overall, this research bridges a gap in Edge AI literature by providing a scalable solution for heterogeneous systems, with potential impacts on 6G standardization and edge computing frameworks. The empirical results from 3,000 runs validate the approach, setting a foundation for adaptive offloading in dynamic environments.

6. Future Applications and Directions

Aikace-aikacen gaba sun haɗa da ingantattun abubuwan Metaverse, sa ido na kiwon lafiya na nesa, da gungun jirage marasa matuka masu cin gashin kansu. Jagororin sun haɗa da haɗa koyon tarayya don sirri, yin amfani da yankin hanyar sadarwa na 6G don rabon albarkatun ƙarfi, da faɗaɗa bayanan martaba don haɗa da tsarin ƙirar ƙwaƙwalwar ƙwaƙwalwa.

7. Nassoshi

1. AI-RAN Alliance. (2023). AI-RAN Working Groups. Retrieved from https://ai-ran.org/working-groups/
2. Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks. In IEEE International Conference on Computer Vision (ICCV).
3. Konečný, J., McMahan, H. B., Yu, F. X., Richtárik, P., Suresh, A. T., & Bacon, D. (2016). Federated Learning: Strategies for Improving Communication Efficiency. arXiv preprint arXiv:1610.05492.
4. IEEE Edge Computing Consortium. (2022). Edge Computing Standards and Practices. Retrieved from https://www.ieee.org