E-PoW Yarjejeniya: Haɗa Koyo na AI da Haƙa Ma'adinai a Tsarin 6G

Teburin Abubuwan Ciki

1. Gabatarwa

Haɗakar Fasahar Hankali na Wucin Gadi (AI) da blockchain a cikin tsarin 6G yana gabatar da matsala ta asali game da albarkatun lissafi. Yayin da horarwar AI ke buƙatar ƙarfin lissafi mai yawa, tsattsarkan Proof-of-Work (PoW) na blockchain suna ɓata albarkatun lissafi masu yawa akan wasu matsalolin sirri. Wannan takarda ta gabatar da Evolved-Proof-of-Work (E-PoW), wata sabuwar hanyar yarjejeniya wacce ke haɗa wannan gibi ta hanyar ba da damar amfani da lissafi biyu don horarwar AI da haƙa ma'adinai.

2. Tsarin Fasaha

2.1 Haɗakar Narkar Matrix

Babban ƙirar ƙima ta ta'allaka ne kan amfani da Lissafin Narkar Matrix (MMC), wanda ya ƙunshi kusan kashi 90% na ayyukan horarwar AI a cikin tsarin kamar Google's Tensor Processing Units. Tushen ilmin lissafi ya haɗa MMC cikin aikin haƙa ma'adinai:

Tsattsarkan PoW yana buƙatar nemo nonce kamar haka:

$H(block\_header + nonce) < target$

E-PoW ya gyara wannan don haɗa ayyukan matrix:

$H(block\_header + nonce + f(A \times B)) < target$

Inda $A$ da $B$ su ne matrices daga ayyukan horarwar AI, kuma $f(\cdot)$ aiki ne na canzawa wanda ke canza samfurin matrix zuwa tsari wanda ya dace da hashing.

2.2 Ƙirar E-PoW Algorithm

Yarjejeniyar E-PoW tana aiki ta hanyar aiki mai zurfi wanda ke kiyaye tsaron blockchain yayin ba da damar sarrafa AI a layi daya. Algorithm ɗin ya tabbatar cewa masu haƙa ma'adinai suna ba da gudummawa ga tabbatar da blockchain da horar da samfurin AI ta hanyoyin lissafi da aka ƙera da kyau.

3. Sakamakon Gwaji

Tabbacin gwaji ya nuna cewa E-PoW na iya ceto har zuwa kashi 80% na ƙarfin lissafi daga tsantsar haƙa ma'adinai don horarwar AI a layi daya. Ma'aunin ayyukan ya nuna:

• Ingantaccen lissafi: sau 3.2 idan aka kwatanta da tsattsarkan PoW
• Hanzarin horarwar AI: sau 2.8 mafi sauri haɗuwa
• Tsaron Blockchain: Yana kiyaye matakin tsaro iri ɗaya kamar na asali na PoW
• Amfani da Albarkatu: Kashi 75-80 na lissafin haƙa ma'adinai an sake amfani da shi don AI

Tsarin gwajin ya ƙunshi gwaji tare da gine-ginen jijiyoyin jiki daban-daban ciki har da Multi-Layer Perceptrons (MLP) da Recurrent Neural Networks (RNN) akan daidaitattun bayanai kamar MNIST da CIFAR-10.

4. Aiwar Code

A ƙasa akwai sauƙaƙan aiwar code na yarjejeniyar E-PoW:

class EPoWConsensus:
    def __init__(self, ai_model, blockchain):
        self.ai_model = ai_model
        self.blockchain = blockchain
        self.matrix_pool = []
    
    def mine_block(self, transactions):
        while True:
            # Sami matrices na horarwar AI
            A, B = self.get_training_matrices()
            
            # Yi narkar matrix don horarwar AI
            C = np.dot(A, B)
            
            # Haɗa sakamako cikin aikin haƙa ma'adinai
            block_header = self.create_block_header(transactions)
            nonce = self.find_nonce(block_header, C)
            
            if self.verify_block(block_header, nonce, C):
                return self.create_block(block_header, nonce, C)
    
    def get_training_matrices(self):
        # Dawo da matrices daga jerin horarwar AI
        if not self.matrix_pool:
            self.matrix_pool = self.ai_model.get_training_batch()
        return self.matrix_pool.pop()

5. Ayyukan Gaba

Yarjejeniyar E-PoW ta buɗe wasu hanyoyin ci gaba masu ban sha'awa don ci gaba:

• Haɗakar AI-Blockchain na Gefe: Tura E-PoW a cikin na'urorin gefe na 6G don rarraba horarwar AI
• Haɓaka Koyo na Tarayya: Yin amfani da blockchain don tabbataccen tara samfura a cikin tsarin koyo na tarayya
• Yunƙurin Blockchain Kore: Rage tasirin muhalli na blockchain ta hanyar aiki mai amfani
• Yankin Hanyar Sadarwa na 6G: Rarraba albarkatu mai ƙarfi tsakanin ayyukan AI da blockchain
• Kasuwanni na AI na Ketare: Ƙirƙirar kasuwanni masu rarrabawa don horar da samfurin AI da fassara

6. Nassoshi

1. Wei, Y., An, Z., Leng, S., & Yang, K. (2023). Connecting AI Learning and Blockchain Mining in 6G Systems.
2. Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system.
3. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.
4. Zhu, J.-Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks. ICCV.
5. Google AI Blog. (2021). Tensor Processing Unit Performance Analysis.
6. 6G Research Vision Papers. (2022). IEEE Communications Society.

7. Bincike Mai Zurfi