Skip to main content

문헌 고찰, Literature Review

· 2 min read

문헌 고찰 개념

An evaluation of previous research on your topic.

  • 문헌 고찰(Literature Review)은 내가 연구하는 주제와 관련된 기존 연구들을 평가하는 글이다.

문헌 고찰 목적

"provides background information needed to understand your study
assures your readers that you are familiar with the important research that has been carried out in your area
establishes your study as one link in a chain of research that is developing and enlarging knowledge in your field"
(Weissberg & Buker 1990, p.41)

  • 내 연구의 배경을 설명하고, 독자에게 내가 주요 연구를 충분히 파악하고 있다는 점, 그리고 내 연구가 기존 연구의 연장선상에 있음을 보여준다.

문헌 고찰 포함 내용

Minimum: well-established research in the field; most recent relevant research.

  • 문헌 고찰에는 반드시 이 분야에서 잘 알려진 주요 연구가장 최근의 관련 연구를 포함해야 한다.

문헌 고찰 구성

  • 필요에 따라 조합할 수 있다.

레퍼런스

The REFERENCE system you follow will be that of one of the leading journals in your field: check with your lecturer or supervisor.

  • 분야에서 영향력 있는 저널의 인용 방식을 따라야 하며, 확실하지 않으면 담당 교수나 슈퍼바이저에게 확인

인용

  • 정보에 초점을 둘지, 저자에 초점을 둘지에 따라 다르다.
  • on the information: use name/number in parentheses.
    • It was demonstrated [2]
    • It was demonstrated (Williams, 2003)
  • on the author: use acknowledging phrases
    • Williams (2003) demonstrated that...
    • Williams [2] demonstrated that...

길이

Around 15-30% of the whole proposal

  • 별도 기준이 없다면 15% - 30%
  • 논문은 60%가 본인의 작업이어야 한다. 문헌 고찰이 논문의 40%를 초과하면 안 된다.

문헌 고찰에서 해야 할 일

  • 분야 내에서 중요하고 관련 있는 연구들을 파악하는 능력을 보여준다.
  • 해당 연구들을 체계적으로 정리, 평가할 수 있음을 보여준다.
  • 연구의 빈틈(gap) 을 파악하고, 내 연구가 그 부분을 채우려 한다는 점을 제시한다.

FDA +001

· 2 min read

Big Data

  • Volume: describes the data size, up to PB
  • Velocity: the speed of data generation and illustrates how fast the datasets are being produced.
  • Variety: the complexity of datasets, with datasets including both structured and unstructured data.
  • Veracity: The truthfulness or reliability of the data. large + reliable in order to achieve value in the analysis of it.
  • Value: The worth in information that can be achieved by the processing and analysis of large datasets. It can be measured by an assessment of the other qualities of big data.

ML

  • Supervised learning: uses the old data to learn the relationship between the inputs and outputs, which will be able to predict the outcomes.
    • Decision Trees, Neural Networks, SVM, Random Forests.
  • Unsupervised learning: try to make sense of the data without labelled responses by representing it in another form.
    • Clustering, association rule mining, Neural Networks.

AI

  • reasoning
  • knowledge representation
  • planning
  • learning
  • natural language processing
  • perception
  • pattern recognition

The Relationships between DA, DS, AI, and ML

Data Science (Second Edition), Chapter 1 - Introduction Pages 1-18, by Vijay Kotu, Bala Deshpande, 2019,

KNIME

KNIME Flow

Vocabulary for AI +002

· 5 min read

Vocabulary & Expressions

Term/ExpressionDefinitionSimpler ParaphraseMeaning
perceiveTo become aware of something through the sensesTo notice or become aware of인식하다
crudeIn a raw or unrefined state; lacking sophisticationBasic or rough; not polished조잡한, 대충만든
flesh outTo add more details or substance to an idea or planTo develop or expand an idea with more information구체화하다, 자세히 설명하다
infraredRelating to electromagnetic radiation with wavelengths longer than visible lightA type of light that is not visible to the human eye, often used in technology적외선의
In practiceIn reality or in actual use, as opposed to theoryHow something works in real life, not just in theory실제로, 실제 사용에서
perceptA mental impression or understanding formed by the sensesAn idea or understanding formed by what we see, hear, etc.지각, 인식
percept sequenceThe complete history of everything the agent has ever perceivedThe full record of all the things an agent has noticed or experienced지각 순서, 인식의 연속
tabulateTo arrange data in a table format for easier analysisTo organize information into a table표로 정리하다
artifactAn object made or used by humans, often of historical or cultural interestA human-made object, especially one of historical or cultural significance인공물, 유물
consequentialismAn ethical theory that judges actions by their outcomes or consequencesA moral philosophy that evaluates actions based on their results결과주의
desirabilityThe quality of being desirable or wantedHow much something is wanted or valued바람직함, 선호도
knottyComplex and difficult to solve or understandComplicated and challenging복잡한, 어려운
criterionA standard or principle by which something is judged or decidedA rule or standard used to evaluate something기준, 표준
priorExisting or occurring before something elseSomething that comes before or is earlier in time이전의, 선행하는
oscillateTo move back and forth in a regular rhythm; to vary between two pointsTo swing or fluctuate between two states or values진동하다, 오락가락하다
omniscienceThe state of having complete or unlimited knowledgeThe ability to know everything전지전능, 모든 것을 아는 상태
pantomimeA form of theatrical performance using gestures and body movements without spoken wordsA performance that tells a story through actions and expressions, not speech무언극, 몸짓극
waspA type of stinging insect known for its slender body and aggressive behaviorA flying insect that can sting, often yellow and black말벌
entomologistA scientist who studies insectsAn expert in the study of insects곤충학자
innateExisting naturally or as part of something; not learned or acquiredSomething that is inborn or inherent타고난, 선천적인
To the extentAs far as something is true or applicableTo the degree that something is relevant or valid~의 범위 내에서, ~에 한해서, ~한 정도까지는
reflexAn automatic, involuntary response to a stimulusA quick, instinctive reaction to something반사 작용, 반사 신경
acronymicallyIn a way that uses the first letters of words to form an abbreviationUsing the initial letters of words to create a shorter form약어로, 두문자어로
aspireTo have a strong desire or ambition to achieve somethingTo aim for or seek to achieve a goal열망하다, 갈망하다
disturbanceAn interruption or disruption of normal activity or stateA disturbance is something that disrupts or interrupts방해, 소란
alleyA narrow passageway between buildings or wallsA small street or path, often for pedestrians골목, 좁은 길
speedometerAn instrument that measures the speed of a vehicleA device that shows how fast a vehicle is going속도계
accelerometerAn instrument that measures acceleration or changes in speedA device that detects how quickly something speeds up or slows down가속도계
persistentContinuing firmly or obstinately in a course of action despite difficulty or oppositionContinuing to exist or endure over a long period지속적인, 끈질긴
embodyTo give tangible or visible form to an idea, quality, or feelingTo represent or express something in a physical form구체화하다, 구현하다
explicitlyIn a clear and detailed manner, leaving no room for confusionClearly stated or shown, without ambiguity명확하게, 분명히
dauntingIntimidating or discouraging; causing one to lose confidenceMaking someone feel less confident or more afraid벅찬, 위협적인, 겁먹게 하는

Agents

  • Simple reflex agents
  • Model-based reflex agents
  • Goal-based agents
  • Utility-based agents

호주 집 구하기

· One min read

시리즈

집 구성

구분영어내용
마스터 룸Master room한국 안방, 화장실이 딸려있어 다른 방에 비해 비싼 편이고 보통 2~4명까지 들어감
세컨드 룸Second room한국 작은방, 2~3명 정도 들어감
리빙 룸Living room거실에서 파티션을 치고 생활하는 것, 가격이 저렴하지만 불편함이 많음.
선 룸Sun room베란다, 베란다여서 저렴할 것 같지만 독방으로 생각하는 사람들도 있어 의외로 가격이 비쌈
본드 피Bond fee한국 월세 보증금, 호주는 주당 방비를 내기 때문에 처음 들어갈 때 4주 정도의 본드 피를 내고 나올 때 돌려받음
풀리 퍼니시드Fully furnished풀 옵션(full option), 침대, 책상, 옷장 등의 가구가 갖춰진 방
어드밴스드 렌트Advanced rent방값 선불

FSD +001

· 8 min read

Software Engineering

SDLC

SW Development Methodologies

Waterfall

  • Pros
    • Simple & easy: each phase has specific deliverables
    • Clear & set milestones
    • Fixed requirements
    • Works well for small projects with specific set of requirements
    • Determine the schedule early
    • Clear structure
  • Cons
    • Working sw produced only at the end
    • High uncertainty of sw quality and functionality
    • Delayed testing, delays sw bugs discovery
    • After completion, no formal way to change the requirements
    • Fix working model, difficult to implement for complex projects

Agile

  • Iterative cyclical progression of the SDLC
    • Repetitive structure based on iterations (sprints)
    • 2 ~ 4 weeks, SDLC repeats
    • Each release has 3 or more iterations
    • Working prototype is produced at the end of each iteration
    • Prototype is for QA and used as input for next iteration
    • Multiple releases
  • Pros
    • Innovation through team collaboration
    • Time to market
    • Continuous testing
    • Risk reduction, finding sw bugs early
    • Customer feedback loop
    • Flexibility to change requirements
    • Automates most of the SDLC === DevOps approach
  • Cons
    • Lack of long-term planning
    • Cost estimation is difficult
    • Limited documentation
    • No finite end
    • Difficult to see the end result due due to cyclic nature of agile

SW Development Paradigms

Procedural Programming

  • Top-down decomposition: Each sub-problem is typically implemented as a function or procedure
  • Root represents the main program
  • Leaves denote individual procedures or functions

Object-Oriented Programming

  • Abstraction: The process of hiding unnecessary details
  • Encapsulation: Restricts the direct access to components of an object, while using methods to access and modify the data
  • Inheritance: The process of creating sub-class
  • Polymorphism: Allows for the creation, use, and storage of multiple objects that inherit from the same parent class

Requirements Analysis

Instructions provided by the stakeholder describe a target system

  • system properties, attributes and how a system should behave
  • Decomposed into Functional requirements / Non-functional requirements

Requirements Validation

  • Compliance
  • Correctness
  • Completeness
  • Consistency
  • Usability
CriteriaDescriptionSatisfactory Score (0-5)Recommendations
ComplianceDegree to which the requirements meet with industry standards and regulations0-
CorrectnessDegree to which the requirements is correct in terms of spelling, accuracy, grammatically.0-
CompletenessDegree to which the functional requirements match the intended software behavior0-
ConsistencyDegree to which the requirements can be mapped to use cases0-
ExpandabilityDegree to which the requirements can be modified and improved to meet the project objectives0-

Use Case

an actor wants the system to, and captures functional requirements

  • always started by an actor, always written from the perspective of the actor
  • a series of actions that a user must initiate to carry out some useful work and to achieve a goal
    • preconditions -> main flow -> alternative flow -> postconditions
  • reflects all the possible events in the system
  • complete set of use cases describes all the possible ways the system will behave and defines all the requirements

Use Case Backlog Template

IDUse Case TitleActor(s)Goal/DescriptionPreconditionsPostconditionsPriorityStatusNotes
UC-001User LoginUserAllow users to log in to the systemUser must have an accountUser is logged inHighOpen-
UC-002View ProfileUserAllow users to view their profile informationUser must be logged inUser profile is displayedMediumOpen-

User Stories

  • Planned to be delivered in a single iteration, some user stories may span multiple iterations
  • Single requirement expressed from developer's perspective
  • Describes a functional or non-functional requirement
  • As a student
  • I want to submit assignments online
  • so that I can receive feedback from my instructor.
IDUser StoryAcceptance CriteriaPriorityStatusStory PointsNotes
US-001As a student, I want to submit assignments online so that I can receive feedback from my instructor.- User can upload assignment files
- User receives confirmation of submission
- Instructor can view submitted assignments
HighOpen5-
US-002As a student, I want to view my grades online so that I can track my academic progress.- User can view grades for each assignment
- User can see overall course grade
- Grades are updated in real-time
MediumOpen3-

UML

  • Actors: Entity that performs actions in the system
  • Use Cases: Oval representation inside the system boundary of a functional requirement
  • System Boundary: Square representation of the system scope
  • Relationships
    • Association: Between an actor and a use case
    • Include: Between use cases, The included use case is always necessary for the completion of the activating use case.
    • Extend: Between use cases, The extension use case is activated occasionally at specific extension point.
    • Generalization/Inheritance: Between use cases, they achieve the same goal but in different ways.

Use Case Model

  • System boundary: Defines the scope of the system
  • Generalization: Describes the shared parts in a parent use case, then specializes in child use cases
    • Inherit features from their parent use case
    • Add new features
    • Change inherited features
  • Short, simple use cases on the core functionality may be completed within one iteration.
  • Keep use cases short and simple
  • Focus on what, not the how
  • Avoid functional decomposition

UML Use Case Model

  • CRUD
    • Create a record
    • Retrieve the record given a key
    • Update the record with new data and store it
    • Delete a record
  • Each is a separate goal, possibly carried by a different person with a different security level.

AI 에이전트와 환경

· 4 min read

에이전트와 환경의 상호작용

환경

  • Static environment: crossword puzzle
  • Semi-dynamic environment: Chess
  • Dynamic environment: self-driving car

연구문제(Research Problem)의 개념, 유형, 선택 전략

· 3 min read

1. 연구문제 정의 및 역할

연구문제는 사회·행동과학에서 분석의 출발점이자 글쓰기의 핵심 조직 원리이며 학문적 소통의 중심 주제이며, 다른 연구 주제로의 확장 기반이 된다.

  • 새로운 지식과 이해 도출
  • 실용적 해결책 또는 정책 제안
  • 인간 조건 개선
  • 실천적 방법론 개발

2. 연구문제 유형과 예시 (기후변화 주제 기반)

유형설명기후 변화 예시
Action사회 변화 유도 목적가뭄 지역 식량 운송 개선
Applied실용적 해결 목표해수면 상승 → 기반시설 개선
Causal인과관계 분석빙하 감소 → 농업 ↓ → 경제 ↓
Developmental개념 확산/수용 과정 분석재생에너지 수용 인식 변화
Exploratory미지 영역 탐색산불 연기 → 노인 건강 영향 조사
Predictive미래 예측북극 해빙 → 항로 개방 대응 시나리오
Theoretical개념적 이론 확장장소애착 이론을 통한 메시지 전략

3. 연구문제 선정 절차

A. 주제를 제공받는 경우

  1. 핵심 개념 및 용어 도출
  2. 배경문헌 검토 (ProQuest, JSTOR 등)
  3. 비판적 시각 형성 (찬반 근거 확보)
  4. 아웃라인 구성

B. 주제 리스트에서 선택하는 경우

  • “쉬운 주제”보다 흥미, 전공, 개인적 의미 고려
  • 위와 같은 절차 적용

C. 자유 주제 설정

  1. “무엇을 이해하고 싶은가?”라는 질문에서 시작
  2. 아이디어 수집 전략
    • 강의자료, 시사 뉴스, SNS, 과거 과제 등
  3. 주제를 연구문제로 구체화 후 위 절차 적용

4. 문헌 활용 전략

활용 방식설명
비판적 활용기존 주장의 약점을 반박하거나 대안 제시
아이디어 확장타인의 관점으로 본인 주장을 보완
역사적 맥락 제공주제의 시간적 변화나 발전 과정을 조망
학제 간 통찰 확보사회학, 정치학 등 다양한 관점 비교 분석

5. 연구문제 vs 논제 (Thesis Statement)

구분연구문제논제 (Thesis)
목적해결해야 할 문제 제시문제에 대한 명확한 입장 표명
문서 위치서론 초반서론 마지막 문단
성격측정 가능한 현상 기술분석, 주장, 평가 등 입장 제시
관계논제의 기반이 됨연구문제 해석 결과로 도출됨

6. 실전 리서치 팁

  • Google Scholar의 "Cited by" 기능으로 후속 연구 추적
  • 동의어 및 관련 키워드로 검색 확장
  • 참고문헌 추적: 자주 인용된 자료는 핵심 이론일 가능성 높음
  • EndNote, RefWorks로 문헌 정리
  • 아웃라인 작성 시 논리적 공백이 생기면 추가 조사 필요

7. 참고용 리서치 도구 및 플랫폼

  • CQ Researcher – 이슈 중심의 정책 보고서 제공
  • Opposing Viewpoints In Context – 사회 문제에 대한 다양한 관점
  • Policy Commons – 글로벌 정책 데이터베이스
  • JSTOR – 학술지 기반의 과거 연구 자료
  • New York Times Topics – 주요 시사 이슈 정리 및 자료

8. 개선 제안 요약

항목제안 내용
유형 비교 도식화7가지 유형을 목적/적용성과 함께 도식화
검색 예시 보강고급 검색식 예시 추가 (ex. “EU AND terrorism AND credibility”)
논제 예시 확장분석형, 설명형, 주장형 등 다양한 논제 샘플 제시
실패 사례 제시과도하게 좁은 주제나 불명확한 문제 사례 소개
전체 개요 문단 추가전체 문서 요약용 300자 내외 개요 삽입 (블로그/논문 개요에 활용 가능)

Vocabulary for AI +001

· 6 min read

Vocabulary & Expressions

Term/ExpressionDefinitionSimpler ParaphraseMeaning
rationalityThe quality of being based on reason or logicLogical thinking합리성
merelyJust; Onlyonly단지
concedeTo admit something is true after denying itadmit인정하다
empiricalBased on observation or experience rather than theoryobservational경험적인
disparageTo regard or represent as being of little worthbelittle폄하하다
extrapolateTo extend the application of a method or conclusion to an unknown situation by assuming existing trends will continueinfer추론하다
devote sth to sth/soTo give one's time, effort, or resources to a particular task or purposededicate헌신하다
preciseMarked by exactness and accuracy of expression or detailexact정확한
contentPleased and satisfied with situationsatisfied만족한
subjectA person or thing being studied or experimented onparticipant피실험자
encyclopediaA book or set of books giving information on many subjects or on many aspects of one subjectreference book백과사전
incorporateTo include or integrate a part into the wholeinclude포함하다
ascertainTo find out or learn with certaintydiscover확인하다
irrefutableImpossible to deny or disproveundeniable반박할 수 없는
syllogismA form of reasoning in which a conclusion is drawn from two given or assumed propositions (premises)logical argument삼단논법
logicianA person skilled in the principles of reasoningreasoner논리학자
seldomAlmost neverrarely드물게
perceptual informationInformation that is derived from the senses and interpreted by the brainsensory information지각 정보
deduceto reach an answer or a decision by thinking carefully about the known factsinfer추론하다
deliberationLong and careful consideration or discussioncareful thought심사숙고
amenableOpen and responsive to suggestion; easily persuaded or controlledopen to suggestion적합한, 순종적인
pervasiveSpreading widely throughout an area or a group of peoplewidespread만연한
a prioriKnowledge or justification that is independent of experienceindependent of experience선험적인
viableCapable of working successfully; feasiblefeasible실행 가능한
soleBeing the only one of its kind; uniqueonly유일한
conjectureAn opinion or conclusion formed on the basis of incomplete informationguess추측
be indicative ofTo be a sign or indication of somethingsuggest~을 나타내다, 잘 보여주다
precursorA person or thing that comes before another of the same kind; a forerunnerpredecessor전신
axiomA statement or proposition that is regarded as being self-evidently trueself-evident truth공리
sufficeTo be enough or adequatebe enough충분하다
deductiveBased on reasoning from the general to the specificlogical reasoning연역적인
coextensiveReach the same extent or scopeof the same extent동등한 범위의
intractabilityThe quality of being difficult to manage or controlunmanageable다루기 어려움, 난해함
appreciationRecognition and enjoyment of the good qualities of someone or somethingrecognition감상, 인식
dampenedMade less intense or strongweakened약화된
confinedRestricted in area or scope; limitedrestricted제한된
gripA firm hold or control over somethingcontrol통제, 지배
animosityStrong hostilityhostility적대감
dwindleTo diminish gradually in size, amount, or strengthdecrease줄어들다
resurgenceAn increase or revival after a period of little activity, popularity, or occurrencerevival부활
spectroscopyThe study of the interaction between matter and electromagnetic radiationstudy of light-matter interaction분광학
assureTo tell someone something positively to dispel any doubtsguarantee보장하다
reinstateTo restore to a previous position or conditionrestore복원하다
extravagantLacking restraint in spending money or using resourceswasteful사치스러운
axiomaticSelf-evident or unquestionableself-evident자명한
brittlenessThe quality of being easily broken or damagedfragility취약성
rebellionAn act of violent or open resistance to an established government or ruleruprising반란
reunificationThe process of reuniting or bringing together againreunification재통일
resurgenceAn increase or revival after a period of little activity, popularity, or occurrencerevival부활
speculationThe forming of a theory or conjecture without firm evidenceguesswork추측
conveneTo come together or assemble, especially for a meetinggather모이다
junctureA point in time, especially one made critical by a concurrence of circumstancescritical point중대한 시점
hinderTo create difficulties for someone or something, resulting in a delay or obstructionobstruct방해하다
sentimentA view or opinion that is held or expressedopinion감정, 의견
reportedlyAccording to what some say; allegedlyallegedly전해진 바에 의하면
prostateA gland in the male reproductive system that produces seminal fluidgland전립선
maneuverTo move skillfully or carefullynavigate조종하다
legged locomotionMovement using legs, such as walking or runningwalking다리로 걷기
upendedTurned upside down or inside outoverturned뒤집힌
adequateSatisfactory or acceptable in quality or quantitysufficient충분한
synthesisThe combination of components or elements to form a connected wholecombination합성
metastaticRelating to the spread of disease from one organ or part to another not directly connected with itspreading전이성의
ophthalmicRelating to the eyeeye-related안과의
facilitateTo make an action or process easy or easierease용이하게 하다

IELTS Transcription @011

· 3 min read

Task1

The line graph below shows the percentage of households with internet access in three countries (Canada, India, and Nigeria) from 2015 to 2020.

task1

The line graph illustrates the proportion of households with internet access in Canada, India, and Nigeria from 2015 to 2020. Overall, all three countries experienced an upward trend in internet access, with Canada maintaining the highest levels, while Nigeria had the lowest throughout the period.

In 2015, Canada had the highest percentage of households with internet access at 80%, rising steadily to 90% by 2020. India started at 20% in 2015 and saw a consistent increase, reaching 45% by 2020. Nigeria, with the lowest initial access at 10%, grew to 25% over the same period.

Canada's internet penetration was significantly higher than that of India and Nigeria, reflecting its advanced infrastructure. India's growth rate was notable, doubling its percentage over five years, indicating rapid digital expansion. Nigeria, while showing progress, remained considerably behind, likely due to economic and infrastructural constraints. These trends highlight disparities in technological development across the countries, with Canada leading and Nigeria lagging.

Task2

The increasing use of social media among young people has raised concerns about its impact on mental health. What are the main causes of this issue, and what measures can be taken to address it?

The growing prevalence of social media among young people has sparked concerns about its detrimental effects on mental health. This essay will explore the primary causes of this issue and propose effective solutions to mitigate its impact.

One major cause is the pressure to conform to idealized online personas. Social media platforms often showcase curated lifestyles, leading to comparison and feelings of inadequacy among young users. For instance, exposure to influencers' polished images can trigger low self-esteem or anxiety. Additionally, excessive screen time, particularly late at night, disrupts sleep patterns, exacerbating stress and metal fatigue. Cyberbullying is another significant factor, as negative comments or harassment online can profoundly affect emotional well-being, especially among adolescents.

To address these issues, educational initiatives should be implemented to promote digital literacy. Schools could teach young people to critically evaluate social media content and recognize its artificial nature, reducing the urge to compare. For example, programs in Australia have successfully integrated media literacy into curricula, fostering healthier online habits. Governments and platforms should also enforce stricter anti-cyberbullying measures, such as real-time moderation and reporting systems. Furthermore, parents and schools can encourage balanced screen time by promoting offline activities, such as sports or hobbies, to enhance mental resilience.

In conclusion, social media's impact on young people's mental health stems from comparison, excessive use, and cyberbullying. By fostering digital literacy, strengthening platform regulations, and encouraging offline engagement, societies can alleviate these effects, ensuring young people maintain healthier relationships with social media.

Vocabulary & Expressions

Term/ExpressionDefinitionSimpler Paraphrase
prominenceThe state of being important or noticeableImportance
conformTo comply with rules or standardsFollow
detrimentalCausing harm or damageHarmful
exacerbatingMaking a problem or situation worseWorsening
resilienceThe ability to recover quickly from difficultiesStrength
alleviateTo make something less severe or intenseReduce
penetrationThe extent to which something is adopted or spreadSpread
disparitiesSignificant differences or inequalitiesGaps

AI의 역사

· 2 min read

A rational agent is one that acts so as to achieve the best outcome or, when there is uncertainty, the best expected outcome

The inception of artificial intelligence (1943–1956)

  • Warren McCulloch와 Walter Pitts (1943): 인공 신경망(artificial neuron model)을 제안.
  • Alan Turing (1950): "Computing Machinery and Intelligence" 논문에서 튜링 테스트(Turing Test)라는 개념을 제안.
  • John McCarthy, Marvin Minsky, Nathaniel Rochester, Claude Shannon이 1956년 Dartmouth workshop을 개최, 여기서 "artificial intelligence"라는 용어가 공식적으로 등장.
  • 주요 성과
    • Logic Theorist (Newell & Simon): 최초의 ‘논리 추론’ 프로그램.
    • General Problem Solver (GPS): 범용 문제 해결 시스템.

Early enthusiasm, great expectations (1952–1969)

  • AI 연구자들은 symbolic reasoning과 search를 이용해 문제를 해결하려 시도.
  • Perceptron(Rosenblatt, 1957): 초기 신경망 모델로, 패턴 인식에 사용.
  • 대표적인 시스템
    • Samuel의 체커 프로그램: 컴퓨터가 스스로 학습해서 체커를 둘 수 있도록 설계.
    • ELIZA: 자연어 처리를 흉내 낸 최초의 대화형 프로그램.

A dose of reality (1966–1973)

  • 자동 번역(Machine Translation)의 한계가 드러남: 1966년 ALPAC 보고서가 정부 자금 축소로 이어짐.
  • Perceptron의 한계(Minsky & Papert, 1969): XOR 문제 등, 단층 신경망으로는 복잡한 문제를 해결할 수 없다는 것을 증명.
  • 연구의 중심이 "toy domains"(작은 문제 영역)로 한정.

Expert systems (1969–1986)

  • Weak methods(약한 방법): 범용성은 있으나 복잡한 문제엔 비효율적.
  • Domain-specific knowledge(도메인 특화 지식): 전문가의 경험을 수천 개의 규칙으로 시스템에 구현.
  • DENDRAL: 질량 분석 데이터를 바탕으로 분자 구조를 추론하는 시스템. Knowledge-intensive system의 대표 사례.
  • MYCIN: 450여 개의 규칙로 혈액 감염을 진단. certainty factors(불확실성 계산) 개념 도입.
  • R1(XCON): Digital Equipment Corporation의 컴퓨터 시스템 주문을 자동으로 구성.
  • 이 시기 Expert system의 상업적 성공: 대기업들이 적극 도입.
  • Frame, Semantic Networks, Description Logics 등 다양한 representation and reasoning tools의 등장.
  • Prolog와 PLANNER와 같은 논리 프로그래밍 언어의 활용.
  • "Fifth Generation" project(일본, 1981): 병렬형 인공지능 컴퓨터 연구 대형 프로젝트(성공적 성과는 제한적).
  • 미국, 유럽도 대응 프로젝트 진행.

The return of neural networks and machine learning (1986–present)

  • Back-propagation 알고리즘이 재발견되며 심층 신경망(deep neural networks)의 부흥이 시작됨.
  • Big Data와 컴퓨팅 파워의 증가로 deep learning이 현실의 다양한 문제를 해결하는 데 쓰이기 시작.
  • 2010년대 Yann LeCun, Geoffrey Hinton, Yoshua Bengio 등이 딥러닝 혁신을 주도.