Tsarin AI Mai Rarraba don Sanya, Zaɓi da Karkatar da Sabis a Cikin Hanyoyin Sadarwar Wayar Hannu

Teburin Abubuwan Ciki

1. Gabatarwa

Haɓakar saurin sabis na AI, musamman manyan samfura kamar jerin GPT na OpenAI, yana canza yanayin trafik a cikin hanyoyin sadarwa na zamani. Duk da yake a halin yanzu manyan kamfanoni ne ke ba da sabis na AI, hasashe suna nuna sauyi zuwa tsarin AI mai rarrabawa inda ƙananan ƙungiyoyi har ma da masu amfani da kansu za su iya ɗaukar nasu samfuran AI. Wannan juyin halitta yana gabatar da manyan ƙalubale wajen daidaita ingancin sabis da jinkiri yayin ɗaukar motsi na mai amfani a cikin kowane tsarin cibiyar sadarwa.

Hanyoyin Lissafin Gefe na Wayar Hannu na Al'ada sun gaza a cikin wannan mahallin saboda dogaro ga tsarin sarrafa matsayi da zato game da hanyoyin sadarwa maras motsi. Girman girman samfuran AI (misali GPT-4 tare da kusan tiriliyan 1.8 na sigogi) ya sa ƙaura na ainihin lokaci ba zai yiwu ba, yana buƙatar ingantaccen mafita don tallafin motsi ba tare da tsadar canja samfura ba.

2. Tsarin Tsarin da Tsarin Matsala

2.1 Samfurin Cibiyar Sadarwa da Abubuwan Da aka Haɗa

Tsarin da aka tsara yana aiki a cikin yanayin cibiyar sadarwa daban-daban wanda ya ƙunshi sabar gajimare, tashoshin tushe, raka'o'in gefen hanya, da masu amfani na wayar hannu. Cibiyar sadarwa tana goyan bayan samfuran AI da aka riga aka horar da su tare da siffofi daban-daban na inganci da jinkiri. Manyan abubuwan da aka haɗa sun haɗa da:

• Sabar Gajimare: Suna ɗaukar manyan samfuran AI tare da babban ƙarfin lissafi
• Tashoshin Tushe & Raka'o'in Gefen Hanya: Suna ba da ɗaukar hoto mara waya da albarkatun lissafin gefe
• Masu Amfani na Wayar Hannu: Suna samar da buƙatun don sabis na AI tare da tsarin motsi
• Samfuran AI: Samfuran da aka riga aka horar da su tare da ma'auni daban-daban na daidaito da jinkiri

2.2 Tsarin Matsala

Matsalar ingantawa ta haɗin gwiwa tana magance matsayin sanyawa, zaɓi, da yanke shawara na karkatarwa don daidaita ingancin sabis da jinkiri har zuwa ƙarshe. Tsarin ya yi la'akari da:

• Jinkirin layin da ba na layi ba a cikin hanyoyin sadarwa
• Tsarin motsi na mai amfani da abubuwan miƙa mulki
• Ƙuntatawar sanyawa na samfuri saboda ƙarancin ajiya
• Bukatun ingancin sabis don aikace-aikace daban-daban

3. Hanyar Fasaha

3.1 Huda Trafik don Tallafin Motsi

Don magance ƙalubalen motsi na mai amfani ba tare da tsadar ƙaura na samfurin AI ba, muna amfani da huda trafik. Lokacin da mai amfani ya motsa tsakanin wuraren shiga mara waya, ainihin wurin shiga yana aiki azaman anga. Amsoshi daga sabar nesa ana mayar da su zuwa wannan hanyar sadarwa ta anga, wanda sai ya tura sakamako zuwa sabon wurin mai amfani. Wannan hanyar:

• Tana kawar da buƙatar ƙaura na samfurin AI na ainihin lokaci
• Tana kiyaye ci gaban sabis yayin abubuwan motsi
• Tana gabatar da ƙarin kayan aikin trafik wanda dole ne a sarrafa su

3.2 Algorithm na Frank-Wolfe Mai Rarraba

Mun ƙirƙira algorithm na ingantawa mai rarrabawa dangane da hanyar Frank-Wolfe tare da sabuwar yarjejeniya ta saƙo. Algorithm:

• Yana aiki ba tare da haɗin kai ba
• Yana jujjuya zuwa ga mafi kyawun yanayi na matsalar da ba ta da daidaituwa
• Yana amfani da iyakantaccen watsa saƙo tsakanin hanyoyin sadarwa maƙwabta
• Yana daidaitawa da canje-canjen yanayin cibiyar sadarwa da buƙatun masu amfani

3.3 Tsarin Lissafi

An tsara matsalar ingantawa a matsayin shiri mara daidaituwa wanda ke la'akari da ma'auni tsakanin ingancin sabis $Q$ da jinkiri har zuwa ƙarshe $L$. Aikin haɗin gwiwa ya haɗu da waɗannan abubuwan:

$$\min_{x,y,r} \sum_{u \in U} \left[ \alpha L_u(x,y,r) - \beta Q_u(x,y) \right]$$

Ƙarƙashin:

$$\sum_{m \in M} s_m y_{n,m} \leq S_n, \forall n \in N$$

$$\sum_{m \in M} x_{u,m} = 1, \forall u \in U$$

$$x_{u,m}, y_{n,m} \in \{0,1\}, r_{u,n} \geq 0$$

Inda $x_{u,m}$ ke nuna mai amfani $u$ ya zaɓi samfuri $m$, $y_{n,m}$ ke nuna hanyar sadarwa $n$ yana ɗaukar samfuri $m$, $r_{u,n}$ shine yanke shawara na karkatarwa, $s_m$ girman samfuri ne, kuma $S_n$ ƙarfin ajiyar hanyar sadarwa ne.

4. Sakamakon Gwaji

4.1 Kimanta Aiki

Ƙididdiga na lissafi sun nuna gagarumin ci gaban aiki akan hanyoyin da suka wanzu. Hanyar da aka tsara tana rage jinkiri har zuwa ƙarshe da kashi 25-40% idan aka kwatanta da hanyoyin warware MEC na al'ada yayin kiyaye ingancin sabis mai kama. Manyan binciken sun haɗa da:

• Huda trafik yana tallafawa motsi yadda ya kamata tare da ƙarancin lalacewar aiki
• Algorithm mai rarrabawa yana da ma'auni yadda ya kamata tare da girman cibiyar sadarwa
• Ingantaccen haɗin gwiwa ya fi hanyoyin yanke shawara na jeri

4.2 Kwatancen da Hanyoyin Tushe

An kwatanta tsarin da aka tsara da hanyoyin tushe guda uku:

• MEC Mai Haɗin Kai: Lissafin gefe na al'ada mai matsayi
• Sanyawa Maras Motsi: Kafaffen sanyawa na samfuri ba tare da daidaitawa ba
• Zaɓin Mai Haɗari: Zaɓin sabis na gani ba tare da haɗin kai ba

Sakamako sun nuna hanyarmu ta sami jinkiri mai ƙasa da kashi 30% fiye da MEC mai haɗin kai da kuma ci gaban kashi 45% akan sanyawa maras motsi a cikin yanayi na babban motsi.

5. Cikakkun Bayanai na Aiwatarwa

5.1 Aiwatar da Lambar

An aiwatar da algorithm na Frank-Wolfe mai rarrabawa tare da waɗannan manyan abubuwan da aka haɗa:

class DecentralizedAIOptimizer:
    def __init__(self, network_graph, models, users):
        self.graph = network_graph
        self.models = models
        self.users = users
        self.placement = {}
        self.routing = {}
        
    def frank_wolfe_iteration(self):
        # Lissafa gradients a cikin gida a kowane hanyar sadarwa
        gradients = self.compute_local_gradients()
        
        # Musayar bayanan gradient tare da maƙwabta
        self.exchange_gradients(gradients)
        
        # Warware matsalar layin gida
        direction = self.solve_linear_subproblem()
        
        # Lissafa girman mataki da sabon mafita
        step_size = self.line_search(direction)
        self.update_solution(direction, step_size)
        
    def optimize(self, max_iterations=100):
        for iteration in range(max_iterations):
            self.frank_wolfe_iteration()
            if self.convergence_check():
                break
        return self.placement, self.routing

5.2 Yarjejeniyar Saƙonni

Sabuwar yarjejeniyar saƙo tana ba da damar haɗin kai mai inganci tsakanin hanyoyin sadarwa tare da ƙaramin kayan aikin sadarwa. Kowane saƙo yana ɗauke da:

• Bayanin gradient na gida don ingantawa
• Matsayin sanyawa da yanke shawara na karkatarwa na yanzu
• Yanayin cibiyar sadarwa da samun albarkatu
• Hasashen motsi na mai amfani

6. Ayyuka na Gaba da Jagorori

Tsarin da aka tsara yana da faffadan aikace-aikace a cikin hanyoyin sadarwa na AI masu tasowa:

• Motocin Cin Gashin Kansu: Ƙaddamar da AI na ainihin lokaci don kewayawa da fahimta
• Birane Masu Hikima: Rarraba sabis na AI don ababen more rayuwa na birane
• IoT na Masana'antu: AI na gefe don masana'antu da kiyayewa na hasashe
• Aikace-aikacen AR/VR: Sarrafa AI maras jinkiri don gogewar nutsawa

Jagororin bincike na gaba sun haɗa da:

• Haɗa kai tare da koyo na tarayya don AI mai kiyaye sirri
• Daidaitawa ga algorithms na ingantawa masu wahayi na ƙididdiga
• Ƙaddamarwa zuwa sabis na AI masu yawa da ingantaccen samfuri
• Haɗa la'akari da ingancin kuzari

7. Bincike na Asali

Wannan bincike yana wakiltar ci gaba mai mahimmanci a cikin sarrafa sabis na AI mai rarrabawa, yana magance manyan ƙalubale a mahadar hanyoyin sadarwar wayar hannu da na'urorin mai kwakwalwa. Sabon amfani da tsarin da aka tsara na huda trafik don tallafin motsi ba tare da ƙaura na samfuri ba yana da mahimmanci musamman, saboda yana kewaye da iyaka na asali na hanyoyin MEC na al'ada lokacin da ake ma'amala da manyan samfuran AI. Kamar yadda CycleGAN (Zhu et al., 2017) ya kawo sauyi ga fassarar hoto zuwa hoto ba tare da bayanan horo biyu ba, wannan aikin yana canza sarrafa motsi a cikin hanyoyin sadarwa masu ba da sabis na AI ta hanyar guje wa aikin ƙaura na samfurin na ainihin lokaci mai tsadar lissafi.

Tsarin lissafi wanda ya haɗa da jinkirin layin da ba na layi ba yana nuna rikitaccen gaskiyar yanayin cibiyar sadarwa, yana motsawa bayan sauƙaƙan samfuran layin da aka saba amfani da su a cikin aikin da ya gabata. Wannan hanyar ta yi daidai da sabbin abubuwan da suka faru a cikin binciken ingantaccen cibiyar sadarwa, kamar aikin Chen et al. (2022) akan lissafin cibiyar sadarwa mara layi, amma ya ƙaddara shi zuwa takamaiman mahallin isar da sabis na AI. Algorithm na Frank-Wolfe mai rarrabawa ya nuna yadda za a iya daidaita dabarun ingantawa na gargajiya zuwa tsarin rarrabawa na zamani, kama da ci gaban kwanan nan a cikin ingantaccen tarayya (Konečný et al., 2016) amma tare da takamaiman daidaitawa ga matsalar sanyawa, zaɓi, da karkatarwa na haɗin gwiwa.

Daga mahangar aiki, ci gaban aikin da aka nuna a cikin sakamakon gwaji (ragewar jinkiri 25-40%) yana da girma kuma zai iya yin tasiri a duniyar gaske akan aikace-aikacen da ke buƙatar ƙaddamar da AI maras jinkiri, kamar motocin cin gashin kansu da sarrafa masana'antu. Kwatancen da hanyoyin tushe yana nuna iyakokin hanyoyin da suke akwai, musamman rashin iyawarsu don magance ƙalubalen musamman da manyan samfuran AI da motsi na mai amfani suka haifar lokaci guda.

Idan aka duba gaba, wannan bincike ya buɗe jagorori masu ban sha'awa da yawa. Haɗin kai tare da fasahohin da ke tasowa kamar hanyoyin sadarwa na 6G da sadarwa ta tauraron dan adam na iya ƙara haɓaka dacewar tsarin. Bugu da ƙari, kamar yadda aka lura a cikin binciken IEEE na kwanan nan game da hankali na gefe, haɓakar bambancin samfuran AI da na'urori masu hanzarin kayan aiki yana gabatar da ƙalubale da dama don ingantaccen rarrabawa. Ƙa'idodin da aka kafa a cikin wannan aikin na iya ba da labari ga ci gaban hanyoyin sadarwa na AI na gaba waɗanda ke haɗa sadarwa, lissafi, da hankali sosai.

8. Nassoshi

1. Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision.
2. Chen, L., Liu, Y., & Zhang, B. (2022). Nonlinear network calculus: Theory and applications to service guarantee analysis. IEEE Transactions on Information Theory.
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. Mao, Y., You, C., Zhang, J., Huang, K., & Letaief, K. B. (2017). A survey on mobile edge computing: The communication perspective. IEEE Communications Surveys & Tutorials.
5. Wang, X., Han, Y., Leung, V. C., Niyato, D., Yan, X., & Chen, X. (2020). Convergence of edge computing and deep learning: A comprehensive survey. IEEE Communications Surveys & Tutorials.
6. Zhang, J., Vlaski, S., & Leung, K. (2023). Decentralized AI Service Placement, Selection and Routing in Mobile Networks. Imperial College London.