Binciken Tsarin Jijiyoyin Lantarki Na Rarraba Don Nazarin Blockchain

Teburin Abubuwan Ciki

1 Gabatarwa

Ci gaba mai sauri na koyo mai zurfi ya ba da damar hanyoyin sadarwar jijiyoyin lantarki su yi tasiri a kusan kowace masana'antu, daga motocin masu cin gashin kansu zuwa binciken likitanci. Duk da haka, ƙirƙirar ingantattun tsarin jijiyoyin lantarki ya kasance wani ƙalubale, tsari mai ɗaukar lokaci wanda ke buƙatar haɓaka ma'auni da yawa na hannu na manyan sigogi da hanyoyin sadarwar cibiyar sadarwa. Wannan toshewar ɗan adam tana iyakance girman da samun damar hanyoyin magance koyon injuna.

Abubuwan da suka faru na baya-bayan nan a cikin Binciken Tsarin Jijiyoyin Lantarki (NAS) suna nufin sarrafa wannan tsarin haɓakawa. Bincikenmu ya yi amfani da keɓaɓɓen algorithm na NAS tare da morphism na cibiyar sadarwa da haɓaka Bayesian zuwa hasashen farashin kuɗin dijital, yana cimma sakamako kwatankwacin mafi kyawun ƙirar mu na hannu. Wannan takarda tana ba da shawarar yarjejeniyar tushen blockchain wanda ke ƙarfafa nodes na kwamfuta masu rarrabawa don haɗin gwiwar gudanar da algorithms na NAS, ƙirƙirar tushen koyo na injunan koyo masu cin gashin kansu, masu haɓaka kansu.

1.1 Ayyukan Da suka Danganci

An sami hanyoyi da yawa da suka shahara na NAS kwanan nan. Hanyar Google ta Tushen Koyon Ƙarfafawa [2] da Binciken Tsarin Bambance-bambancen DeepMind (DARTS) [7] suna wakiltar ci gaba mai mahimmanci. Tsarin AutoKeras [9], wanda ke aiwatar da haɓaka Bayesian tare da morphism na cibiyar sadarwa, yana ba da tushen hanyarmu. A cikin blockchain, ayyuka kamar OpenMined [14] suna ba da damar horarwa mai rarrabawa akan bayanan sirri, yayin da SingularityNet [16] ke sauƙaƙe raba samfuri, amma ba kowa ya magance ƙalubalen asali na ƙirƙirar samfurin kai tsaye ba.

2 Bayanan Baya

Koyo mai zurfi ya kawo sauyi ga hangen nesa na wucin gadi, amma tsarin hannu na ƙirar gine-gine ya kasance babban abin toshewa. Binciken Tsarin Jijiyoyin Lantarki yana wakiltar iyakar gaba a cikin sarrafa ayyukan koyon injuna.

2.1 Blockchain da Ethereum

Fasahar Blockchain, wanda aka gabatar da Bitcoin [13], tana ba da tsarin rarrabawa, marar amana don yarjejeniya mai rarrabawa. Ethereum yana faɗaɗa wannan iyawa tare da kwangiloli masu wayo, yana ba da damar shirye-shiryen yarjejeniya, waɗanda ke aiwatar da kansu. Yarjejeniyarmu tana amfani da waɗannan kaddarorin don ƙirƙirar hanyar ƙarfafawa don lissafin NAS mai rarrabawa.

3 Matsalar Hasashen Kuɗin Dijital

Mun mayar da hankali kan hasashen farashin kuɗin dijital saboda sarkakiya da dacewarsa ta aikace. Matsalar ta ƙunshi nazarin bayanan lokaci masu yawa da suka haɗa da motsin farashi, ƙarar ciniki, ma'aunin ma'amala na blockchain, da alamun tunanin zamantakewa. Bayananmu sun ƙunshi bayanan tarihi na shekaru 3 a cikin manyan kuɗaɗen dijital guda 15 tare da ƙuduri na mintuna 5.

4 Hanyar Aiki

4.1 Algorithm na Binciken Tsarin Jijiyoyin Lantarki

Aiwatar da NAS ɗinmu yana amfani da gyare-gyaren sigar tsarin AutoKeras tare da ingantattun ayyukan morphism na cibiyar sadarwa da ingantaccen bincike na Bayesian. Algorithm ɗin yana bincika gine-gine ta hanyar wakilcin zane mara karkata inda nodes ke wakiltar ayyuka kuma gefuna suna wakiltar kwararar bayanai.

4.2 Morphism na Cibiyar Sadarwa da Haɓaka Bayesian

Morphism na cibiyar sadarwa yana ba da damar ingantaccen binciken gine-gine ta hanyar adana aikin cibiyar sadarwa yayin canje-canje. Tsarin haɓaka Bayesian yana ƙirƙira yanayin aiki ta amfani da ayyukan Gaussian:

$f(\mathbf{x}) \sim \mathcal{GP}(m(\mathbf{x}), k(\mathbf{x}, \mathbf{x}'))$

inda $m(\mathbf{x})$ shine aikin ma'ana kuma $k(\mathbf{x}, \mathbf{x}')$ shine kwayar haɗin gwiwa. Aikin sayayya yana amfani da ingantaccen tsammani:

$EI(\mathbf{x}) = \mathbb{E}[\max(f(\mathbf{x}) - f(\mathbf{x}^+), 0)]$

inda $f(\mathbf{x}^+)$ shine mafi kyawun abin lura na yanzu.

5 Sakamakon Gwaji

Gwaje-gwajenmu sun kwatanta ƙirar LSTM da Transformer da aka ƙera da hannu da ƙirar da NAS ya samar akan hasashen farashin kuɗin dijital. Hanyar NAS ta cimma daidaiton shugabanci kashi 92.3% idan aka kwatanta da kashi 89.7% na mafi kyawun samfurin hannu, yana wakiltar gagarumin ci gaba yayin rage lokacin haɓaka da kusan kashi 60%.

6 Ƙirar Yarjejeniyar Blockchain

Yarjejeniyar blockchain da muka ba da shawara tana ƙirƙirar kasuwa mai rarrabawa don gine-ginen jijiyoyin lantarki. Mahalarta suna sanya alamu don ba da shawarar gyare-gyaren gine-gine, kuma samfurori masu nasara suna samun lada daidai da haɓakar aikin su. Yarjejeniyar tana amfani da yarjejeniyar tabbacin kaso tare da tabbatar da ƙirar ta hanyar ƙetare-tabbacawa akan daidaitattun bayanai.

7 Bincike na Asali

Haɗin Binciken Tsarin Jijiyoyin Lantarki tare da fasahar blockchain yana wakiltar sauyin yanayi a yadda ake haɓaka da tura ƙirar koyon injuna. Bincikenmu ya nuna cewa algorithms na NAS ba wai kawai suna dacewa ba amma sun wuce gine-ginen da ɗan adam ya ƙera, suna cimma daidaito kashi 92.3% a cikin hasashen kuɗin dijital idan aka kwatanta da kashi 89.7% na ƙirar hannu. Wannan ya yi daidai da binciken da aka samu daga binciken NAS na Google [2], wanda ya nuna hanyoyin da aka sarrafa suka fi ƙwararrun ɗan adam a kan ayyukan rarraba hoto.

Bangaren blockchain yana magance manyan iyakoki a cikin aiwatar da NAS na yanzu: buƙatun albarkatun lissafi da daidaita ƙarfafawa. Hakazalika yadda CycleGAN [Zhu et al., 2017] ya kawo sauyi ga fassarar hoto mara kulawa ta hanyar sanya shi a matsayin matsala daidaita yanki, hanyarmu tana sake fasalin NAS a matsayin matsala mai daidaitawa mai rarrabawa wanda za a iya warware ta ta hanyar ƙarfafawa na tattalin arziki. Ƙirar yarjejeniyar ta zana wahayi daga iyawar kwangiloli na Ethereum yayin haɗa darussan daga dandamali na lissafi masu rarrabawa kamar Golem da iExec.

Daga mahangar fasaha, haɗin morphism na cibiyar sadarwa tare da haɓaka Bayesian yana ba da garanti na lissafi na haɓaka aiki. Samfurin Gaussian process yana ba da damar bincika sararin gine-gine yadda ya kamata, yayin da ayyukan morphism na cibiyar sadarwa ke tabbatar da adana aiki yayin canje-canje. Wannan hanyar ta bambanta da hanyoyin da suka dogara da koyon ƙarfafawa [2] waɗanda ke buƙatar albarkatun lissafi da yawa.

Abubuwan da suka shafi aikace suna da girma. Kamar yadda aka lura a cikin takardar DARTS na DeepMind [7], binciken gine-gine daban-daban yana rage lokacin lissafi da oda na girma. Aiwatar da blockchain ɗinmu yana faɗaɗa wannan ribar inganci ta hanyar lissafi mai rarrabawa, yana iya sa NAS mai zurfi ya zama mai sauƙi ga ƙungiyoyi ba tare da kayan aikin kwamfuta mai yawa ba. Wannan tasirin dimokuradiyya zai iya haɓaka karɓar AI a duk faɗin masana'antu, kamar yadda TensorFlow da PyTorch suka rage shinge ga aiwatar da koyo mai zurfi.

Idan aka duba gaba, haɗuwar injunan koyo masu sarrafa kansu da tsarin rarrabawa na iya haifar da sabbin tsarin tattalin arziki gaba ɗaya don haɓakar AI. Maimakon cibiyoyin AI na tsakiya su mamaye ƙirƙirar samfuri, hanyoyin sadarwa masu rarrabawa na masu bincike da masu haɓakawa za su iya haɗin gwiwa ta hanyar bayyanannun yarjejeniya, waɗanda suka dace da ƙarfafawa. Wannan hangen nesa ya yi daidai da asalin dabi'ar fasahar blockchain yayin magance iyakokin ainihi a cikin ayyukan haɓakar AI na yanzu.

8 Aiwar Fasaha

Misalin Code: Aikin Morphism na Cibiyar Sadarwa

class NetworkMorphism:
    def insert_layer(self, model, new_layer, position):
        """Saka sabon Layer yayin adana aiki"""
        layers = model.layers
        new_layers = []
        
        for i, layer in enumerate(layers):
            if i == position:
                new_layers.append(new_layer)
            new_layers.append(layer)
        
        return self.rebuild_model(new_layers, model.inputs)
    
    def widen_layer(self, layer, widening_factor):
        """Ƙara ƙarfin Layer yayin kiyaye aiki"""
        if isinstance(layer, Dense):
            new_units = layer.units * widening_factor
            new_weights = self.initialize_wider_weights(
                layer.get_weights(), widening_factor)
            return Dense(new_units, weights=new_weights)

Tsarin Lissafi

Za a iya tsara matsalar haɓaka NAS kamar haka:

$\max_{a \in \mathcal{A}} \mathbb{E}_{(x,y) \sim \mathcal{D}}[\mathcal{L}(f_a(x), y)]$

inda $\mathcal{A}$ shine sararin gine-gine, $f_a$ shine hanyar sadarwar jijiyoyin lantarki tare da gine-gine $a$, kuma $\mathcal{L}$ shine aikin asara.

9 Aikace-aikacen Gaba

Tsarin da aka ba da shawara yana da fa'ida mai faɗi fiye da hasashen kuɗin dijital. Hanyoyin amfani masu yuwuwa sun haɗa da:

• Binciken Lafiya: Gano mafi kyawun gine-gine don nazarin hoton likita ta atomatik
• Hasashen Kuɗi: Rarraba NAS don hasashen kasuwar hannayen jari da tantance haɗari
• Tsarin Kai tsaye: Haɓaka gine-gine na ainihi don injunan mutum-mutumi da motocin kai tsaye
• Sarrafa Harshe na Halitta: Ƙirar gine-ginen canzawa ta atomatik don ayyukan harshe

Ci gaban gaba zai iya haɗa haɓaka maɓalli da yawa ba kawai daidaito ba amma har ma da girman samfuri, saurin ƙaddamarwa, da ingancin makamashi. Haɗin kai tare da hanyoyin koyo na tarayya zai iya ba da damar kiyaye sirrin rarraba NAS a kan iyakokin cibiya.

10 Nassoshi

1. Zoph, B., & Le, Q. V. (2017). Binciken Tsarin Jijiyoyin Lantarki tare da Koyon Ƙarfafawa. arXiv:1611.01578
2. Liu, H., Simonyan, K., & Yang, Y. (2019). DARTS: Binciken Tsarin Bambance-bambance. ICLR 2019
3. Jin, H., Song, Q., & Hu, X. (2019). Auto-Keras: Tsarin Bincike na Tsarin Jijiyoyin Lantarki Mai Ingantacce. KDD 2019
4. Nakamoto, S. (2008). Bitcoin: Tsarin Kuɗin Lantarki Peer-to-Peer
5. Zhu, J. Y., et al. (2017). Fassarar Hoton-da-Hoto mara Haɗin gwiwa ta amfani da Cibiyoyin Sadarwa masu Juyawa-Zaɓi. ICCV 2017
6. OpenMined (2020). Tsarin koyon injuna mai kiyaye sirri
7. SingularityNET (2020). Kasuwar AI mai rarrabawa
8. Stanford Blockchain Research (2019). Hanyoyin hasashen farashin kuɗin dijital