Daidaitawa (Alignment)
Alignment
Tsarin tabbatar da cewa manufofin tsarin AI, abubuwan da aka samar, da halaye sun dace da manufofin ɗan adam da ƙima. Wannan yana da mahimmanci musamman a cikin tsarin ci gaba waɗanda zasu iya haɓaka halaye waɗanda ba a yi niyya ba.
Misali: Tabbatar da cewa chatbot don tallafin lafiyar hankali baya taɓa ba da shawarar ayyukan cutarwa ba tare da la'akari da umarni ba.
Tsarin Shirye-shiryen Aikace-aikace (API) (Application Programming Interface (API))
Application Programming Interface (API)
Saitin ƙa'idodi da ka'idoji da aka ayyana waɗanda ke ba da damar tsarin software daban-daban su sadarwa da musayar bayanai.
Misali: Amfani da OpenAI API don aika umarni da karɓar amsa da aka samar da samfurin harshe a cikin aikace-aikacen yanar gizon ku.
Hankali Na GabaÉ—aya (AGI) (Artificial General Intelligence (AGI))
Artificial General Intelligence (AGI)
Wani nau'i na AI na ka'ida wanda zai iya yin kowane aikin hankali da ɗan adam zai iya. Yana haɓaka koyo a cikin yankuna daban-daban.
Misali: Tsarin AGI zai iya koyon tsara kiÉ—a, yin tiyata, da cin jarabawar falsafa ba tare da shirye-shiryen takamaiman aiki ba.
Hankali Na Artificial (AI) (Artificial Intelligence (AI))
Artificial Intelligence (AI)
Kwaikwayon hankalin É—an adam a cikin injuna waÉ—anda aka shirya don tunani, dalili, da aiki da kansu.
Misali: AI tana ba da ikon mataimakan sirri kamar Siri da tsarin tuƙi mai cin gashin kai kamar Tesla Autopilot.
ÆŠabi'ar AI (AI Ethics)
AI Ethics
Wani fanni da ya shafi tasirin ɗabi'a na haɓaka da amfani da AI, gami da adalci, sirri, alhaki, da rashin nuna bambanci.
Misali: Ƙirƙirar jagororin don hana algorithms na haya daga nuna bambanci dangane da jinsi ko kabila.
Hankali Mai Haɓaka (Augmented Intelligence)
Augmented Intelligence
Samfurin haɗin gwiwa inda AI ke haɓaka da haɓaka hankalin ɗan adam maimakon maye gurbinsa.
Misali: Kayan aikin rediyo mai amfani da AI waɗanda ke nuna abubuwan da ba a saba gani ba ga likitoci, waɗanda ke yin ganewar ƙarshe.
Wakili Mai Cin Gashin Kai (Autonomous Agent)
Autonomous Agent
Tsarin AI mai iya yanke shawararsa da É—aukar matakai don cimma burinsa ba tare da sa hannun É—an adam ba.
Misali: Robot mai tuƙi da kansa yana tafiya titunan birni kuma yana guje wa cikas da kansa.
Komawa Baya (Backpropagation)
Backpropagation
Hanyar horar da cibiyoyin sadarwa na jijiyoyi ta hanyar sabunta nauyi a baya daga fitarwa zuwa yadudduka na shigarwa, rage kurakuran hasashe.
Misali: Ana amfani da shi wajen horar da masu rarraba hotuna don rage yawan kuskure wajen gane lambobi da aka rubuta da hannu.
Son Zuciya (Algorithmic Bias) (Bias (Algorithmic Bias))
Bias (Algorithmic Bias)
Son zuciya mara niyya da tsari a cikin sakamakon AI saboda bayanan horo marasa daidaito ko marasa wakilci.
Misali: Tsarin gane fuska wanda ke gano mutane masu launi da yawa saboda rashin wakilci a cikin bayanan horo.
Manyan Bayanai (Big Data)
Big Data
Manyan bayanai masu girma waɗanda ke buƙatar kayan aiki na musamman don adanawa, nazari, da fitar da ƙima, galibi ana amfani da su don horar da samfuran AI.
Misali: Amfani da miliyoyin mu'amalar mai amfani don horar da injunan ba da shawarwari don dandamali na e-commerce.
Samfurin Akwatin Baƙi (Black Box Model)
Black Box Model
Wani nau'i na AI ko samfurin koyon inji wanda ba a iya fassara tsarin ciki cikin sauƙi ga ɗan adam, yana sa ya zama da wahala a fahimci yadda ake yanke shawara.
Misali: Cibiyar sadarwa ta jijiyoyi mai zurfi da aka yi amfani da ita don amincewa da lamuni amma ba ta ba da bayani bayyananne game da dalilin da ya sa aka karɓi mai nema ɗaya kuma aka ƙi wani.
Kwamfuta Mai Fahimta (Cognitive Computing)
Cognitive Computing
Tsarin AI da aka tsara don kwaikwayon tsarin tunanin É—an adam, kamar dalili da koyo, ta amfani da dabaru kamar NLP da gane tsari.
Misali: Tsarin kwamfuta mai fahimta wanda ke taimaka wa ƙwararrun shari'a su nazari dokokin shari'a da hasashen sakamako.
Ganin Kwamfuta (Computer Vision)
Computer Vision
Wani fanni na fasahar hankali wanda ke ba da damar kwamfutoci su fassara da sarrafa bayanan gani kamar hotuna da bidiyo.
Misali: Tsarin gane fuska waÉ—anda ke gano mutane a cikin hotunan tsaro ta amfani da ganin kwamfuta.
Tarin Rubutu (Corpus)
Corpus
Babban tarin rubuce-rubuce ko maganganu da aka yi amfani da su don horar da samfuran harshe.
Misali: Bayanan Common Crawl tarin rubutu ne na yanar gizo da aka yi amfani da shi don horar da manyan samfuran harshe kamar GPT.
Canjin Bayanai (Data Drift)
Data Drift
Abin da ke faruwa inda bayanan shigarwa ke canzawa a kan lokaci, yana sa aikin samfurin ya lalace.
Misali: Samfurin kiyayewa na hasashe don kayan aikin masana'antu ya zama maras daidaito yayin da aka gabatar da sabon fasahar firikwensin.
Lakabin Bayanai (Data Labelling)
Data Labelling
Tsarin sanya lakabi ko alamomi ga bayanai don sa ya dace da koyon kulawa.
Misali: Sanya lakabi ga dubban hotunan ciwon daji a matsayin marasa lahani ko masu lahani don horar da samfurin gano ciwon daji.
Haƙar Bayanai (Data Mining)
Data Mining
Tsarin gano tsari mai ma'ana, alaƙa, da abubuwan da ba a saba gani ba a cikin manyan bayanai.
Misali: Masu sayarwa suna amfani da haƙar bayanai don gano cewa mutanen da ke siyan diapers galibi suna siyan giya ma.
Koyon Zurfi (Deep Learning)
Deep Learning
Wani fanni na koyon inji wanda ke amfani da cibiyoyin sadarwa na jijiyoyi masu yadudduka da yawa don tsara tsari masu rikitarwa a cikin bayanai.
Misali: Ana amfani da koyon zurfi a cikin samfuran harshe kamar GPT-4 da samfuran samar da hoto kamar Stable Diffusion.
Samfuran YaÉ—uwa (Diffusion Models)
Diffusion Models
Wani nau'i na samfuran samarwa waÉ—anda ke koyon samar da bayanai ta hanyar canza hayaniya bazuwar zuwa abubuwan da aka tsara.
Misali: Stable Diffusion yana ƙirƙirar hotuna masu kama da gaske daga umarnin rubutu ta amfani da dabaru na yaɗuwa.
HaÉ—awa (Embedding)
Embedding
Wakilcin vector na bayanai, galibi ana amfani da shi don kama ma'anar semantic na kalmomi, hotuna, ko jimloli.
Misali: A cikin NLP, kalmar 'banki' na iya samun haÉ—awa iri É—aya da 'kuÉ—i' amma daban da 'bankin kogi' dangane da mahallin.
Zamani (Epoch)
Epoch
Cikakken maimaitawa a kan dukkan bayanan horo yayin tsarin horar da samfurin koyon inji.
Misali: Idan bayanan suna da misalai 1,000 kuma samfurin ya ga dukkan su sau É—aya yayin horo, wannan shine zamani É—aya.
AI Mai ÆŠabi'a (Ethical AI)
Ethical AI
Falsafar ƙira da turawa wanda ke tabbatar da cewa fasahar AI tana aiki a fili, daidai, kuma daidai da ƙimar al'umma.
Misali: Kayan aikin haya na AI wanda ya haÉ—a da binciken son zuciya don hana nuna bambanci ga 'yan takara marasa rinjaye.
Tsarin Kwararru (Expert System)
Expert System
Tsarin AI wanda ke kwaikwayon ikon yanke shawara na ƙwararren ɗan adam a cikin takamaiman yanki ta amfani da ƙa'idodi da dabaru.
Misali: Tsarin ƙwararru da aka yi amfani da shi a aikin gona don ba da shawarar maganin amfanin gona dangane da bayanan ƙasa da tarihin kwari.
AI Mai Bayyanawa (XAI) (Explainable AI (XAI))
Explainable AI (XAI)
Tsarin AI da aka tsara don sa tsarin ciki da yanke shawara su zama masu fahimta ga ɗan adam, yana ƙara amana da alhaki.
Misali: AI na ganewar asali na likita wanda ba kawai yana ba da shawarwari ba har ma yana bayyana waɗanne alamomi ne suka haifar da wannan ƙarshe.
Koyon Ƙananan Harbi (Few-shot Learning)
Few-shot Learning
Hanyar koyon inji inda aka horar da samfurin ko aka gyara shi ta amfani da Æ´an misalai kaÉ—an da aka lakaba.
Misali: Gyara LLM don rubuta imel na shari'a bayan nuna masa misalai 10 kawai.
Gyara (Fine-tuning)
Fine-tuning
Tsarin ɗaukar samfurin da aka riga aka horar da shi da kuma ci gaba da horar da shi a kan sabon, ƙaramin bayanan don ƙware shi don takamaiman aiki.
Misali: Gyara babban LLM kamar GPT a kan takardun shari'a na ciki don ƙirƙirar mataimakin rubutun shari'a.
Samfurin Tushe (Foundation Model)
Foundation Model
Babban samfurin da aka horar da shi a kan bayanai daban-daban da faÉ—i waÉ—anda za a iya daidaita su zuwa ayyuka da yawa.
Misali: GPT-4 da PaLM 2 samfuran tushe ne masu iya taƙaitawa, Q&A, fassara, da ƙari.
Logic Mai RuÉ—i (Fuzzy Logic)
Fuzzy Logic
Wani nau'i na dabaru wanda ke magance ƙimar kusan maimakon ƙayyadaddun dabaru na gaskiya/ƙarya (binary), mai amfani don dalili a ƙarƙashin rashin tabbas.
Misali: Ana amfani da shi a cikin tsarin sarrafa yanayi don daidaita zafin jiki dangane da shigarwar ruÉ—i kamar 'É—an zafi' ko 'sanyi sosai'.
Cibiyar Sadarwa Mai Haɓaka (GAN) (Generative Adversarial Network (GAN))
Generative Adversarial Network (GAN)
Tsarin samfurin samarwa inda cibiyoyin sadarwa guda biyu — mai samarwa da mai rarrabawa — ke gasa don inganta ingancin fitarwa.
Misali: Ana amfani da GANs don ƙirƙirar bidiyo na deepfake ko samar da hotunan samfur masu kama da gaske daga zane-zane.
AI Mai Haɓaka (Generative AI)
Generative AI
Wani nau'i na fasahar hankali wanda zai iya ƙirƙirar sabon abun ciki — kamar rubutu, hotuna, kiɗa, ko bidiyo — daga bayanan horo.
Misali: ChatGPT yana samar da rubutun blog ko Midjourney yana ƙirƙirar fasahar dijital daga umarnin rubutu.
Transformer Mai Haɓaka (GPT) (Generative Pre-trained Transformer (GPT))
Generative Pre-trained Transformer (GPT)
Wani nau'i na manyan samfuran harshe da OpenAI ya haɓaka wanda ke amfani da tsarin transformer kuma an riga an horar da shi a kan manyan bayanai na rubutu don yin ayyuka daban-daban na harshe.
Misali: GPT-4 yana iya rubuta kasidu, fassara harsuna, da taƙaita takardu tare da ɗan ƙaramin umarni.
Algorithm Na Halitta (Genetic Algorithm)
Genetic Algorithm
Hanyar ingantawa da aka yi wahayi zuwa ga zaɓin halitta inda mafita ke haɓaka a kan lokaci ta hanyar maye gurbi, haɗuwa, da zaɓi.
Misali: Ana amfani da shi don ƙirƙirar ingantattun tsarin cibiyar sadarwa ta jijiyoyi ta hanyar kwaikwayon rayuwar mafi dacewa.
RuÉ—u (Hallucination)
Hallucination
Samar da abun ciki mai kama da gaske amma ba daidai ba ko mara ma'ana ta samfurin AI.
Misali: Samfurin harshe yana ƙirƙirar ambato mara wanzuwa ko yana ba da bayanan tarihi na ƙarya.
Heuristic (Heuristic)
Heuristic
Hanyar aiki don warware matsala wanda baya tabbatar da cikakkiyar mafita amma ya isa ga manufofin nan take.
Misali: Amfani da ƙa'idar babban yatsa don kimanta lokacin bayarwa a cikin tsarin AI na dabaru.
Hyperparameter (Hyperparameter)
Hyperparameter
Ƙimar daidaitawa da aka saita kafin horar da samfurin koyon inji, kamar matakin koyo ko adadin yadudduka.
Misali: Daidaita girman batch daga 32 zuwa 128 don inganta saurin horo da aikin samfurin.
Inference (Inference)
Inference
Tsarin amfani da samfurin koyon inji da aka horar don yin hasashe ko samar da abubuwan da aka fitar daga sabbin bayanan shigarwa.
Misali: Amfani da samfurin GPT da aka gyara don rubuta imel ga ƙungiyar tallafin abokin ciniki.
Gano Nufi (Intent Detection)
Intent Detection
Aiki a cikin fahimtar harshe na halitta inda tsarin ke gano burin mai amfani ko manufa a cikin saƙo.
Misali: A cikin chatbot, gane 'Ina so in yi ajiyar jirgi' a matsayin niyyar ajiyar tafiya.
Intanet Na Abubuwa (IoT) (Internet of Things (IoT))
Internet of Things (IoT)
Cibiyar sadarwa ta na'urorin jiki masu haÉ—in kai da aka saka da firikwensin, software, da sauran fasahohi don tattarawa da musayar bayanai.
Misali: Smart thermostats da firiji waÉ—anda ke ba da rahoton bayanan amfani da daidaita saituna ta amfani da nazarin AI.
Fassarawa (Interpretability)
Interpretability
Yadda É—an adam zai iya fahimtar tsarin ciki na samfurin koyon inji da tsarin yanke shawararsa.
Misali: Itacen yanke shawara ya fi fahimta fiye da cibiyar sadarwa ta jijiyoyi mai zurfi saboda yanke shawararsa ana iya gano su.
Jupyter Notebook (Jupyter Notebook)
Jupyter Notebook
Yanayin kwamfuta mai ma'amala mai buɗaɗɗen tushe wanda ke ba masu amfani damar rubuta lambar, nuna abubuwan da aka fitar, da rubuta nazari a cikin keɓaɓɓen keɓaɓɓen.
Misali: Masana kimiyyar bayanai suna amfani da Jupyter Notebooks don tsara samfuran koyon inji da raba sakamako.
Maƙwabta Mafi Kusa (KNN) (K-Nearest Neighbours (KNN))
K-Nearest Neighbours (KNN)
Algorithm na koyon inji mai sauƙi, mara ma'ana da aka yi amfani da shi don rarrabawa da regression. Yana yanke shawara dangane da misalan horo mafi kusa a cikin sararin fasali.
Misali: Don rarraba sabon 'ya'yan itace a matsayin apple ko pear, KNN yana bincika waÉ—anne 'ya'yan itatuwa da aka lakaba ne mafi kusa a siffa da launi.
Taswirar Ilimi (Knowledge Graph)
Knowledge Graph
Tsarin bayanai wanda ke amfani da nodes da gefuna don wakiltar da adana bayanan da aka haɗa na abubuwa da alaƙarsu.
Misali: Panel É—in ilimi na Google yana amfani da taswirar ilimi wanda ke haÉ—a abubuwa kamar mutane, wurare, da abubuwan da suka faru.
Inganta Samfurin Koyon Harshe (LLMO) (Language Learning Model Optimisation (LLMO))
Language Learning Model Optimisation (LLMO)
Dabaru da aka yi amfani da su don inganta aiki, inganci, ko daidaitawa na manyan samfuran harshe don takamaiman ayyuka ko yankuna.
Misali: Amfani da quantisation da daidaita umarni don inganta LLM don amfani da kasuwanci.
Babban Samfurin Harshe (LLM) (Large Language Model (LLM))
Large Language Model (LLM)
Wani nau'i na samfurin koyon zurfi da aka horar da shi a kan manyan bayanai na rubutu masu iya samarwa, fahimta, da dalili tare da harshen É—an adam.
Misali: ChatGPT da Claude LLMs ne da aka horar don taimakawa wajen rubutu, coding, da amsa tambayoyi.
Sararin Boye (Latent Space)
Latent Space
Wakilcin abstract mai girma inda aka haÉ—a shigarwa iri É—aya kusa da juna, ana amfani da shi a cikin samfuran samarwa da haÉ—awa.
Misali: A cikin samar da hoto, sarrafa sararin boye zai iya canza fasali kamar haske ko motsin rai.
Matsayin Koyon (Learning Rate)
Learning Rate
Babban hyperparameter a cikin horo wanda ke sarrafa yadda ake daidaita nauyin samfurin dangane da gradient na asara.
Misali: Babban matakin koyo na iya haifar da wuce gona da iri, yayin da ƙaramin matakin koyo ke rage saurin ci gaban horo.
Koyon Inji (ML) (Machine Learning (ML))
Machine Learning (ML)
Wani reshe na AI wanda ke ba da damar tsarin su koyi daga bayanai da inganta aiki ba tare da an shirya su a fili ba.
Misali: Tacewar spam suna amfani da koyon inji don rarraba imel a matsayin spam ko a'a dangane da misalan da suka gabata.
Canjin Samfurin (Model Drift)
Model Drift
Abin da ke faruwa inda daidaiton samfurin ke raguwa a kan lokaci saboda canje-canje a cikin bayanai ko yanayi.
Misali: Samfurin gano zamba ya zama maras daidaito yayin da dabaru na zamba ke haɓaka.
Koyarwar Samfurin (Model Training)
Model Training
Tsarin ciyar da bayanai zuwa samfurin koyon inji da daidaita sigoginsa don rage kuskure.
Misali: Horar da injin ba da shawarwari a kan tarihin sayan abokin ciniki don ba da shawarar sabbin samfurori.
AI Mai Harsuna Da Yawa (Multimodal AI)
Multimodal AI
Tsarin AI mai iya sarrafawa da haÉ—a nau'ikan bayanai daban-daban kamar rubutu, hotuna, sauti, da bidiyo.
Misali: Samfurin kamar GPT-4 Vision wanda zai iya karanta rubutu da fassara hotuna a lokaci guda.
Sarrafa Harshe Na Halitta (NLP) (Natural Language Processing (NLP))
Natural Language Processing (NLP)
Wani fanni na AI da aka mayar da hankali kan mu'amala tsakanin kwamfutoci da harsunan É—an adam (na halitta). Yana ba da damar injuna su karanta, fahimta, da amsa a cikin harshen É—an adam.
Misali: Ana amfani da NLP a cikin mataimakan murya, aikace-aikacen fassarar harshe, da chatbots.
Cibiyar Sadarwa Ta Jijiyoyi (Neural Network)
Neural Network
Samfurin koyon inji da aka yi wahayi zuwa ga tsarin kwakwalwar ɗan adam, wanda ya ƙunshi yadudduka na nodes masu haɗin kai (neurons).
Misali: Cibiyoyin sadarwa na jijiyoyi suna bayan samfuran koyon zurfi da aka yi amfani da su a cikin gane hoto da magana.
Hayaniya (Noise)
Noise
Bayanan bazuwar ko marasa mahimmanci a cikin bayanai waÉ—anda zasu iya É“oye tsari masu ma'ana da kuma shafar aikin samfurin.
Misali: Kurakuran firikwensin ko shigarwar bayanai masu cike da kuskure za a iya É—aukar su a matsayin hayaniya.
Ontology (Ontology)
Ontology
Tsarin tsari wanda ke rarraba da ayyana alaƙa tsakanin ra'ayoyi a cikin yanki, galibi ana amfani da shi a cikin tsarin AI na semantic.
Misali: Ontology a cikin kiwon lafiya na iya ayyana yadda alamomi ke da alaƙa da cututtuka da magani.
Overfitting (Overfitting)
Overfitting
Kuskuren samfurin inda samfurin koyon inji ke kama hayaniya a cikin bayanan horo kuma yana yin aiki mara kyau a kan sabbin bayanai.
Misali: Samfurin da ke haddace amsoshin horo amma ba zai iya sarrafa bayanan gwaji da ba a gani ba an yi masa overfitting.
Nazarin Hasashe (Predictive Analytics)
Predictive Analytics
Amfani da bayanai, algorithms, da AI don gano yiwuwar sakamako na gaba dangane da bayanan tarihi.
Misali: Masu sayarwa suna amfani da nazarin hasashe don hasashen buƙatar wasu samfurori.
Koyarwar Farko (Pre-training)
Pre-training
Tsarin horar da samfurin da farko a kan babban, bayanan gabaÉ—aya kafin gyara shi don takamaiman ayyuka.
Misali: Samfuran GPT an riga an horar da su a kan manyan tarin rubutu kafin a daidaita su don chatbots na sabis na abokin ciniki.
Injiniyan Prompt (Prompt Engineering)
Prompt Engineering
Fasaha da kimiyya na ƙirƙirar umarni masu inganci don sarrafa fitarwa na manyan samfuran harshe.
Misali: Ƙara umarnin tsarin kamar 'Amsa a matsayin malami mai ladabi' misali ne na injiniyan prompt.
Quantisation (Quantisation)
Quantisation
Hanyar matsa lamba ta samfurin wanda ke rage adadin bits da aka yi amfani da su don wakiltar nauyi da kunna, yana haɓaka inganci.
Misali: Quantising samfurin daga 32-bit zuwa 8-bit yana inganta aiki a kan na'urorin hannu.
Kwamfuta Mai Kwantum (Quantum Computing)
Quantum Computing
Sabuwar tsarin kwamfuta dangane da injiniyan kwantum, wanda ke da damar samar da ƙarfin sarrafawa mai yawa.
Misali: Kwamfuta mai kwantum na iya hanzarta horar da AI fiye da iyakokin gargajiya.
Injin Dalili (Reasoning Engine)
Reasoning Engine
Tsarin a cikin AI wanda ke fitar da ƙarshe na dabaru daga saitin gaskiya ko bayanai ta amfani da ƙa'idodi ko algorithms na inference.
Misali: Kayan aikin ganewar asali na AI yana amfani da injin dalili don fitar da yiwuwar yanayin likita dangane da alamomi.
Koyon Ƙarfafawa (RL) (Reinforcement Learning (RL))
Reinforcement Learning (RL)
Wani yanki na koyon inji inda wakilai ke koyo ta hanyar mu'amala da muhallinsu don haɓaka lada mai tarin yawa.
Misali: Robot yana koyon tafiya ta hanyar gwaji da kuskure ta amfani da dabaru na RL.
Koyon Ƙarfafawa Tare Da Ra'ayin Dan Adam (RLHF) (Reinforcement Learning with Human Feedback (RLHF))
Reinforcement Learning with Human Feedback (RLHF)
Hanyar koyo inda zaɓin ɗan adam ke jagorantar siginar lada ta AI, galibi ana amfani da ita wajen gyara samfuran harshe.
Misali: An horar da ChatGPT tare da RLHF don samar da amsoshi masu taimako da aminci.
Samar Da Haɓaka (RAG) (Retrieval-Augmented Generation (RAG))
Retrieval-Augmented Generation (RAG)
Hanyar haÉ—a dawo da bayanai tare da samarwa, inda LLM ke É—aukar takardu masu dacewa don inganta amsarsa.
Misali: Mataimakin AI yana dawo da kuma ambato bayanan samfur yayin samar da amsa ga tambaya ta fasaha.
Koyon Kai-da-Kai (Self-Supervised Learning)
Self-Supervised Learning
Hanyar horo inda samfurin ke koyon tsari ta hanyar samar da lakabinsa daga bayanan da ba a sarrafa ba, yana rage dogaro ga bayanan da É—an adam ya lakaba.
Misali: An horar da BERT tare da koyon kai-da-kai ta hanyar hasashen kalmomin da suka É“ace a cikin rubutu.
Binciken Semantic (Semantic Search)
Semantic Search
Hanyar bincike wanda ke fahimtar niyyar mai amfani da ma'anar mahallin, ba kawai daidaita kalmomi ba.
Misali: Binciken 'yadda ake gyara famfo mai zubarwa' yana dawo da jagororin ko da kalmar 'famfo mai zubarwa' ba ta cikin takardar.
Nazarin Ji (Sentiment Analysis)
Sentiment Analysis
Tsarin gano motsin rai, ra'ayoyi, ko halaye a cikin rubutu, galibi ana rarraba su a matsayin tabbatacce, mara kyau, ko tsaka-tsaki.
Misali: Nazarin tweets don auna martanin jama'a ga sabon samfur.
Stochastic (Stochastic)
Stochastic
Ya haÉ—a da bazuwar ko halayen yiwuwar, galibi ana amfani da shi a cikin AI mai samarwa da algorithms na ingantawa.
Misali: Fitarwa na GPT-4 yana bambanta don shigarwa iri É—aya saboda tsarin sa na stochastic decoding.
AI Mai Ƙarfi (Strong AI)
Strong AI
Wanda aka fi sani da Hankali Na GabaÉ—aya (AGI), yana nufin injuna masu ikon fahimta na matakin É—an adam a duk yankuna.
Misali: AI na gaba wanda zai iya rubuta littattafai da kansa, tsara birane, da warware matsalolin É—abi'a daidai.
Babban Hankali Na Artificial (SAI) (Super Artificial Intelligence (SAI))
Super Artificial Intelligence (SAI)
AI na ka'ida wanda ya zarce hankalin ɗan adam a kowane fanni — dalili, kerawa, hankali na motsin rai, da sauransu.
Misali: SAI zai iya haɓaka sabbin kimiyya da falsafa da kansa.
Koyon Kulawa (Supervised Learning)
Supervised Learning
Hanyar koyon inji inda aka horar da samfuran a kan bayanan da aka lakaba don koyon taswirar shigarwa-fitarwa.
Misali: Koyar da samfurin don rarraba imel a matsayin spam ko a'a ta amfani da misalan tarihi.
Bayanan Halitta (Synthetic Data)
Synthetic Data
Bayanan da aka samar da fasaha wanda ke kwaikwayon bayanan duniya na gaske, galibi ana amfani da shi don horo lokacin da bayanan gaske suka yi ƙaranci ko masu mahimmanci.
Misali: Ƙirƙirar hotunan likita na halitta don horar da samfuran ganewar asali ba tare da keta sirrin mai haƙuri ba.
Token (Token)
Token
Rukunin rubutu da LLMs ke sarrafawa — galibi kalma ko ɓangaren kalma.
Misali: Jimlar 'Sannu duniya!' an raba shi zuwa tokens 3: 'Sannu', 'duniya', da '!'.
Tokenisation (Tokenisation)
Tokenisation
Tsarin raba rubutu zuwa tokens don sarrafawa ta samfurin.
Misali: A cikin NLP, 'ChatGPT yana da kyau' ya zama ['Chat', 'G', 'PT', 'yana', 'da', 'kyau'].
Koyon Canja Wuri (Transfer Learning)
Transfer Learning
Amfani da ilimi daga aiki ɗaya don haɓaka koyo a kan wani aiki mai alaƙa, yana rage lokacin horo da buƙatun bayanai.
Misali: Gyara samfurin da aka horar da shi a kan rubutun Ingilishi don yin nazarin ji a wani harshe.
Transformer (Transformer)
Transformer
Tsarin cibiyar sadarwa ta jijiyoyi wanda ke amfani da hanyoyin kulawa don tsara bayanan jere, ana amfani da shi sosai a cikin LLMs.
Misali: BERT, GPT, da T5 duk samfuran transformer ne.
Underfitting (Underfitting)
Underfitting
Lokacin da samfurin ya yi sauƙi don kama tsari a cikin bayanan horo, yana haifar da aiki mara kyau.
Misali: Samfurin layi wanda ke ƙoƙarin hasashen rarraba hotuna masu rikitarwa na iya yin underfit.
Koyon Ba Tare Da Kulawa Ba (Unsupervised Learning)
Unsupervised Learning
Hanyar koyo inda samfuran ke gano tsari ko tarin a cikin bayanan da ba a lakaba ba.
Misali: Rarraba abokan ciniki dangane da halayen sayan ba tare da lakabi da aka ayyana ba.
Nufin Mai Amfani (User Intent)
User Intent
Burin ko manufa a bayan tambayar mai amfani ko mu'amala.
Misali: Mai amfani da ke rubuta 'yadda ake gasa cake' mai yiwuwa yana nufin nemo girke-girke.
Saitin Tabbatarwa (Validation Set)
Validation Set
Wani É“angare na bayanai da aka yi amfani da shi don kimanta aikin samfurin yayin horo da daidaita hyperparameters.
Misali: Ana amfani da shi don gano overfitting kafin gwajin ƙarshe.
Database Na Vector (Vector Database)
Vector Database
Database da aka tsara don adanawa da bincika haÉ—in vector da aka yi amfani da shi a cikin ayyukan AI kamar binciken kama da RAG.
Misali: Pinecone da Weaviate databases ne na vector don adana rubutu ko haÉ—in hoto.
HaÉ—in Vector (Vector Embedding)
Vector Embedding
Wakilcin lambobi na bayanai wanda ke adana ma'anar semantic da alaƙa a cikin sararin vector.
Misali: Kalmomin 'sarki' da 'sarauniya' suna da haÉ—awa iri É—aya tare da bambance-bambance na jinsi.
Mataimakin Virtual (Virtual Assistant)
Virtual Assistant
Wakili na software mai amfani da AI wanda ke taimaka wa masu amfani su kammala ayyuka ta hanyar tattaunawa ko umarnin murya.
Misali: Siri, Alexa, da Google Assistant mataimakan virtual ne masu shahara.
Gane Murya (Voice Recognition)
Voice Recognition
Fasaha wanda ke fassara da canza harshen magana zuwa rubutu ko aiki.
Misali: Rubutun murya da umarnin murya suna dogara ga tsarin gane murya.
AI Mai Rauni (Weak AI)
Weak AI
Tsarin AI da aka tsara don yin takamaiman aiki ba tare da hankali na gabaÉ—aya ba.
Misali: AI mai kunna chess wanda ba zai iya fahimtar harshe ko tuƙi mota ba misali ne na AI mai rauni.
Web Scraping (Web Scraping)
Web Scraping
Cire bayanai ta atomatik daga gidajen yanar gizo, galibi ana amfani da shi don tattara bayanan horo ko saka idanu kan abun ciki.
Misali: Scraping jerin gidaje don horar da samfurin kimanta kadarori.
Nauyi (Weight)
Weight
Sigar a cikin cibiyoyin sadarwa na jijiyoyi wanda ke ƙayyade ƙarfin tasirin node ɗaya a kan wani.
Misali: Nauyi yana daidaitawa yayin horo don rage kuskuren samfurin.
Whisper (Whisper)
Whisper
Samfurin magana-zuwa-rubutu da OpenAI ya haɓaka mai iya rubuta sauti a cikin harsuna da yawa.
Misali: Whisper zai iya rubuta laccoci da podcasts tare da babban daidaito.
YAML (YAML)
YAML
Tsarin da É—an adam zai iya karantawa don serialisation na bayanai, galibi ana amfani da shi don fayilolin daidaitawa a cikin ayyukan koyon inji.
Misali: Ayyanawa sigogin samfurin a cikin fayil É—in YAML don horo a cikin PyTorch.
Koyon Zero-shot (Zero-shot Learning)
Zero-shot Learning
Ikon samfurin don yin ayyukan da ba a taɓa horar da shi a fili ba ta hanyar amfani da ilimi na gabaɗaya.
Misali: Samfurin yana amsa tambayoyin shari'a duk da cewa ba a horar da shi musamman a kan bayanan shari'a ba.
Zettabyte (Zettabyte)
Zettabyte
Rukunin bayanan dijital daidai da sextillion É—aya (10^21) bytes, galibi ana amfani da shi don bayyana girman bayanan intanet.
Misali: Yawan zirga-zirgar intanet na duniya ya wuce zettabyte 1 a kowace shekara a shekarar 2016.
