What US programmes ask for美国各项目要求
In the US the central document is the statement of purpose, sent with a CV and usually three letters of recommendation; there is normally no research proposal. Universities name it differently (statement of objectives at MIT, academic statement at Northwestern and Michigan, statement of academic purpose at Yale and Princeton, candidate statement at Chicago) and set lengths of about 1-2 pages or up to 1,000 words in most programmes. Each programme sets its own length, tests and deadline, and most PhD deadlines in the data fall between mid-November and mid-December 2026.
在美国,核心文书是目的陈述(SOP),与简历和通常三封推荐信一起提交,一般不需要研究计划。各校名称不同(麻省理工称 statement of objectives,西北大学与密歇根称 academic statement,耶鲁与普林斯顿称 statement of academic purpose,芝加哥称 candidate statement),多数项目篇幅约 1-2 页或不超过 1,000 词。篇幅、考试与截止日期由各项目自行规定,数据中多数博士截止日期在 2026 年 11 月中旬至 12 月中旬之间。
Computer science and engineering programmes ask for a statement of purpose of 1-2 pages (Stanford Electrical Engineering up to 2 pages single-spaced; Carnegie Mellon Computer Science 1-2 pages; Harvard SEAS Computer Science up to 1,000 words; UCLA Computer Science up to 500 words plus a 500-word personal statement; Penn Engineering 2 pages answering six prompts). You usually select faculty in the application rather than email them first: Carnegie Mellon says no adviser is needed before applying, and the U Washington Allen School needs no prior faculty contact. The GRE is optional or not required in most rows (Northwestern Computer Science still requires it).
计算机与工程项目要求 1-2 页的目的陈述(斯坦福电子工程单倍行距不超过 2 页;卡内基梅隆计算机 1-2 页;哈佛 SEAS 计算机不超过 1,000 词;UCLA 计算机不超过 500 词,另加 500 词个人陈述;宾大工程学院 2 页内回答六个问题)。通常在申请表中选择老师,而不是先发邮件:卡内基梅隆说明申请前无须确定导师,华盛顿大学 Allen 学院无须事先联系老师。多数项目 GRE 可选或不要求(西北大学计算机仍要求)。
Science & Engineering: what each programme asks for理工科:各项目要求
| UNIVERSITY 大学 | WHAT THE PROGRAMME ASKS FOR 项目要求 | SUPERVISOR CONTACT 联系导师 | OFFICIAL PAGE 官方页面 |
|---|---|---|---|
| MIT 麻省理工学院 QS 2027: 1 | Statement of objectives; Chemistry also a 500-1,000-word personal statement + 1,000-1,500-word objectives naming 3 faculty Chemistry: GRE not accepted; MechE letters due 1 Nov References: 3 | No; name faculty in the statement | oge.mit.edu |
| Stanford 斯坦福大学 QS 2027: =2 | Electrical Engineering: SOP ≤2 pages single-spaced, naming research interests and faculty References: 3 | No; name faculty in the SOP / form | ee-www.stanford.edu |
| Harvard 哈佛大学 QS 2027: 5 | SEAS Computer Science: SOP ≤1,000 words naming labs PhD deadline 15 Dec References: 3 | Not stated centrally | seas.harvard.edu |
| Caltech 加州理工学院 QS 2027: 7 | SOP ~2 pages + CV ~2 pages (Geological & Planetary Sciences); application not reviewed until all 3 letters arrive Deadlines 1-15 Dec by option References: 3 (max 3) | Not stated centrally | gps.caltech.edu |
| Penn 宾夕法尼亚大学 QS 2027: 15 | Penn Engineering PhD (incl. CIS): personal statement max 2 pages answering six prompts + resume; no writing sample GRE optional; IELTS 7.5 / TOEFL 100 recommended, no minimum; window 15 Sep to 15 Dec 2026 References: 2 (CIS 3; 1+ from faculty) | Not stated centrally | gradadm.engineering.upenn.edu |
| Cornell 康奈尔大学 QS 2027: =16 | Engineering: academic SOP ≤2 pages naming faculty + personal statement MAE 1 Dec; CEE 7 Jan References: 3 | No; English asks you not to contact faculty | engineering.cornell.edu |
| Yale 耶鲁大学 QS 2027: 16 | Chemistry: personal statement of purpose GRE + TOEFL; due 1 Dec References: 3+ | Not stated centrally | chem.yale.edu |
| Johns Hopkins 约翰斯·霍普金斯大学 QS 2027: 20 | Computer Science: statement, letters and grades are the main components All materials incl. letters by 1 Dec | Not stated centrally | cs.jhu.edu |
| UC Berkeley 加州大学伯克利分校 QS 2027: =20 | Mechanical Engineering: SOP 1-2 pages double-spaced + Personal History 1-2 pages (no overlap) References: 3 | Select a faculty adviser in the form (Social Welfare) | me.berkeley.edu |
| Chicago 芝加哥大学 QS 2027: 24 | Mathematics: statement; do not submit GRE IELTS 7.0 / TOEFL 90 References: 3+ | Not stated centrally | mathematics.uchicago.edu |
| Princeton 普林斯顿大学 QS 2027: 27 | Computer Science: no minimum English score for PhD admission Below IELTS 8 speaking → oral test in year 1 | Not stated centrally | cs.princeton.edu |
| Columbia 哥伦比亚大学 QS 2027: =43 | SEAS doctoral (incl. Computer Science): personal statement 250-1,000 words recommended + CV GRE not required (2026 cycle); doctoral deadline 15 Dec References: 3 | Not stated centrally | engineering.columbia.edu |
| Northwestern 西北大学 QS 2027: =45 | Computer Science: academic statement explaining fit with a named adviser; resume; may name up to 2 faculty GRE required; due 1 Dec 2026 References: 2 minimum, 3 recommended | No; Learning Sciences says contact not required, naming 1-3 faculty helps | mccormick.northwestern.edu |
| UCLA 加州大学洛杉矶分校 QS 2027: 49 | Computer Science: statement of purpose ≤500 words + personal statement ≤500 words; no CV required GRE optional; due 15 Dec; IELTS 7.0 References: 3 | No; Psychology lists which faculty accept students | cs.ucla.edu |
| Michigan 密歇根大学 QS 2027: 51 | CSE: statement of purpose (research interests, goals) + personal statement; choose faculty in application GRE not considered References: 3 suggested | No; name matching faculty in the academic statement | cse.engin.umich.edu |
| Carnegie Mellon 卡内基梅隆大学 QS 2027: 55 | Computer Science Dept: statement of purpose 1-2 pages + CV; select faculty in application GRE optional; final deadline 9 Dec 2026; no adviser needed before applying References: 3 | Not required; Psychology strongly encourages contacting faculty | csd.cmu.edu |
| NYU 纽约大学 QS 2027: 58 | Courant Computer Science: statement of purpose + CV GRE not required; due 12 Dec References: 3 | No; name available mentors in the SOP | cs.nyu.edu |
| Brown 布朗大学 QS 2027: 66 | Computer Science: personal statement; research motivation valued; no need to name faculty GRE optional; due 15 Dec 2026 References: 3 | No; Psychology invites contact but it is not required | cs.brown.edu |
| Duke 杜克大学 QS 2027: 70 | Computer Science: statement of purpose + resume; no writing sample GRE optional; due 15 Dec (fall only); TOEFL, IELTS or Duolingo for non-native speakers References: 3 | No; CS says contact is not needed | gradschool.duke.edu |
| UT Austin 得克萨斯大学奥斯汀分校 QS 2027: 72 | Computer Science: statement of purpose (why UT CS) GRE optional; due 15 Dec, letters 20 Dec References: 3 or more | No; Educational Psychology asks you to name a matching faculty member | cs.utexas.edu |
| UIUC 伊利诺伊大学厄巴纳-香槟分校 QS 2027: 74 | Siebel Computer Science: academic statement of purpose 1-2 pages + CV GRE optional; late applications not accepted References: 3 | Not stated centrally | siebelschool.illinois.edu |
| UC San Diego 加州大学圣地亚哥分校 QS 2027: 81 | CSE: statement of purpose (no word limit) + CV No GRE for Fall 2027; due 16 Dec 2026 References: 3 | No; name up to 3 advisers in the application | cse.ucsd.edu |
| Penn State 宾夕法尼亚州立大学 QS 2027: =92 | Computer Science & Engineering: statement of purpose + CV/resume GRE not required for PhD; TOEFL/IELTS per Graduate School (GCAC-305); deadline on CSE admissions site References: 3 | No; select main faculty choice + 2 others | bulletins.psu.edu |
| U Washington 华盛顿大学 QS 2027: =92 | Allen School CSE: no prior faculty contact or chosen adviser needed Deadline not stated (fee requests by ~6 Dec) | No; Allen School needs no prior faculty contact | cs.washington.edu |
| Boston University 波士顿大学 QS 2027: 94 | Computer Science: personal statement + CV/resume GRE not mandatory; due 1 Dec References: 3 | Not stated centrally | bu.edu |
| Purdue 普渡大学 QS 2027: =100 | Computer Science: academic statement of purpose + personal history statement + resume GRE not required; due 1 Dec 2026; IELTS 7.5 recommended | Not stated centrally | cs.purdue.edu |
| University of Wisconsin-Madison 威斯康星大学麦迪逊分校 QS 2027: =131 | Computer Sciences: statement of purpose + resume/CV + supplemental sections GRE not required; due 15 Dec References: 3 | Psychology: admission by individual faculty; check or contact faculty about whether they admit that year | guide.wisc.edu |
| University of California, Davis 加州大学戴维斯分校 QS 2027: 137 | Computer Science: SOP relating plans to UC Davis research + personal history statement + CV Due 15 Dec, no spring entry; min GPA 3.0; TOEFL/IELTS unless English-medium degree References: 3 | Psychology strongly recommends emailing potential mentors before the deadline | cs.ucdavis.edu |
| Georgia Institute of Technology 佐治亚理工学院 QS 2027: =142 | PhD CS: online application; SOP length not stated on page; desirable GPA 3.5 GRE optional; IELTS 7.5 (R/L/S 6.5, W 5.5) or TOEFL 100; due 15 Dec References: 3 (academic preferred) | Psychology encourages contacting aligned faculty before applying | cc.gatech.edu |
| University of Southern California 南加州大学 QS 2027: 153 | Viterbi PhD (incl. CS): personal statement (no length) + resume/CV; research ability expected GRE not required for CS; due 15 Dec 2026 References: 3 (faculty) | Optional; Rossier allows contacting 1 to 2 faculty; Psychology asks you to name 1 to 3 | viterbigradadmission.usc.edu |
| University of North Carolina at Chapel Hill 北卡罗来纳大学教堂山分校 QS 2027: =158 | Computer Science: statement of purpose + resume/CV (guidance on cs.unc.edu) GRE not required; listing shows 9 Mar 2027 deadline (fellowships 15 Dec) References: 3 | Name 1 to 5 faculty mentors in application (Psychology, Epidemiology) | gradschool.sites.unc.edu |
| Washington University in St. Louis 圣路易斯华盛顿大学 QS 2027: =162 | McKelvey Engineering PhD (incl. CS): SOP 1 to 2 pages + CV GRE optional (no effect); IELTS 6.5 / TOEFL 90; deadline on separate page References: 3 | Not stated centrally | engineering.washu.edu |
| Texas A&M University 得克萨斯农工大学 QS 2027: 169 | CSE: one-page SOP on research interests and closely related faculty + CV (min 1 page, research and publications) GRE optional; TOEFL below 100 makes admission unlikely; due 15 Dec References: Not stated (research advisers preferred) | Name faculty in the statement; CSE does not ask for prior contact | engineering.tamu.edu |
| Arizona State University 亚利桑那州立大学 QS 2027: 172 | SCAI Computer Science: SOP about 1 to 2 pages naming 2 to 3 ASU faculty; CV encouraged; min GPA 3.5 IELTS 7 / TOEFL 90; due 31 Dec; GRE policy unclear on page References: 3 | C&I Language and Literacy requires an approved mentor agreeing to advise | scai.engineering.asu.edu |
| University of California, Santa Barbara 加州大学圣塔芭芭拉分校 QS 2027: 173 | Computer Science: SOP + personal history statement (min 250 words, no duplication) + CV IELTS 7 / TOEFL 100; due 15 Dec (page shows 2025 date); GRE not mentioned References: 3 | Not stated centrally | graddiv.ucsb.edu |
| Michigan State University 密歇根州立大学 QS 2027: 182 | Chemistry PhD: academic statement + personal statement/SOP (no lengths) + CV + Chemistry Applicant Datasheet GRE General and Subject recommended, not required; international deadline 15 Dec (page shows Fall 2026 cycle) References: 3 minimum | Psychology recommends contacting faculty whose research matches yours | chemistry.msu.edu |
| Emory University 埃默里大学 QS 2027: 183 | Computer Science and Informatics: SOP (research questions, focus, plans) + CV GRE optional; IELTS 7 / TOEFL 85; due 15 Dec References: 3 | Rank up to 3 faculty mentors (Psychology); name 2+ faculty (Epidemiology) | computerscience.emory.edu |
Requirements are quoted from the official page linked in each row, in English as published. The supervisor column is the university-wide rule. 表中要求按各行所附官方页面的英文原文摘录;“联系导师”一栏为全校通用规则。
37 programme pages, data checked 2026-11-18. Rules change each cycle; the programme page always wins. 共 37 个项目页面,数据核对日期 2026-11-18。规则每年可能调整,以项目页面为准。
The statement of purpose in a science and engineering application理工科申请中的目的陈述
In science and engineering the statement of purpose is short, and research experience carries most of the weight: the problem, the techniques, the code or instruments, and the results. Most engineering and computer science programmes admit you to work with an adviser or ask you to select faculty in the application, while some life-science programmes admit you to the programme and let you choose a lab after rotations. The six-part structure is the same in every field.
理工科的目的陈述篇幅较短,研究经历分量最重:问题、技术、代码或仪器,以及结果。多数工程与计算机项目按导师录取,或要求在申请表中选择老师;部分生命科学项目则按项目录取,第一年轮转后再选实验室。六部分结构在各方向通用。
How the statement changes across the field本领域内不同方向的差异
| FIELD 方向 | WHAT RESEARCH EXPERIENCE SHOWS 研究经历要体现什么 | WHAT FIT MEANS 匹配的含义 |
|---|---|---|
| Electrical · Mechanical engineering 电气 · 机械工程 | Simulations, test rigs, hardware or code you built, with the result in numbers. 你搭建的仿真、试验台、硬件或代码,并用数字写出结果。 | A group whose methods and facilities match the problem, often named in the form. 方法与设施与问题匹配的课题组,常需在申请表中写明。 |
| Computer science 计算机 | Projects, papers and repositories; the model or system, the data and the evaluation. 项目、论文与代码仓库;模型或系统、数据与评估方式。 | Faculty you select in the application, with one link each to your interests. 在申请表中选择的老师,每位对应一处研究兴趣。 |
| Chemistry · Materials 化学 · 材料 | Syntheses, characterisation techniques and instruments you ran yourself. 你亲手完成的合成、表征技术与使用过的仪器。 | Two or three faculty whose labs use or need your techniques. 实验室使用或需要你所掌握技术的两三位老师。 |
| Biology · Life sciences 生物 · 生命科学 | Lab techniques, model systems, months of bench time and your own results. 实验技术、模式系统、实验室工作时长与你自己的结果。 | Labs you would like to rotate in, and why, where the programme uses rotations. 采用轮转制的项目中,你希望轮转的实验室及理由。 |
| Mathematics · Physics 数学 · 物理 | Courses beyond the degree, reading projects, problems you solved and any research output. 超出学位要求的课程、阅读项目、解决过的问题与研究成果。 | An area and the faculty working in it; specific problems where you can name them. 研究领域及该领域的老师;能写出具体问题更好。 |
What a statement of purpose is not目的陈述不是什么
It gives a direction and one or two questions, not a design with a timeline.
它给出研究方向和一两个问题,而不是附时间表的完整设计。
The committee has your CV. Pick two or three projects and say what each taught you.
评审手里有你的简历。挑两三个项目,说明每个项目让你学到了什么。
Background belongs in the second statement, where one is asked for.
个人背景放在第二份陈述里(如有要求)。
Most US universities call it a statement of purpose. Others call it a statement of objectives (MIT), an academic statement (Northwestern, Michigan), a statement of academic purpose (Yale, Princeton, Columbia), a candidate statement (Chicago) or a research statement (Carnegie Mellon, U Washington); Brown, Boston University and USC call the main essay a personal statement. Whatever the name, keep it short and technical: committees in these fields read for what you built and what it showed.
多数美国大学称之为目的陈述(statement of purpose)。也有学校称为目标陈述(麻省理工)、学术陈述(西北大学、密歇根)、学术目的陈述(耶鲁、普林斯顿、哥伦比亚)、候选人陈述(芝加哥)或研究陈述(卡内基梅隆、华盛顿大学);布朗、波士顿大学和南加州大学把主文书称为个人陈述。无论叫什么名称,都要简短、具体、技术化:这些领域的评审看的是你做了什么、得到了什么。
How long, and how the words are shared篇幅与字数分配
Write one full master version of about 1,000 to 1,200 words, then cut it to size. Research experience and research interests get the most words; the opening and closing stay short. Shorter limits keep the same order and cut each part in proportion. Each section page then gives a blueprint, a worked example, a weak versus stronger comparison and lines from other fields. In the US data most limits fall between 500 words (UCLA Computer Science) and 2 pages (Stanford Electrical Engineering, single-spaced), with a few at 3-5 pages (Michigan Ross, U Washington Education).
先写一份约 1,000 至 1,200 词的完整母版,再按要求裁剪。研究经历和研究兴趣字数最多,开篇与结尾保持简短。篇幅更短时,保持同样顺序,各部分按比例压缩。之后每个部分都包含结构说明、范例、“较弱”与“更好”写法的对比,以及其他方向的示例。美国数据中,多数篇幅限制在 500 词(UCLA 计算机)到 2 页(斯坦福电子工程,单倍行距)之间,少数为 3-5 页(密歇根 Ross、华盛顿大学教育学院)。
Cutting to size按篇幅压缩
| LIMIT 限制 | WHAT TO KEEP 保留什么 |
|---|---|
| About 1,000 words or 2 pages 约 1,000 词或 2 页 | The master version: all six sections, two or three projects in research experience. 使用母版:六个部分全部保留,研究经历写两三个项目。 |
| About 500 words or 1 page 约 500 词或 1 页 | One sentence of opening, your strongest project, one question, two named faculty, one sentence of goals. 开篇一句、最强的一个项目、一个问题、两位老师、目标一句。 |
| A character limit or set questions 字符限制或固定问题 | Answer each prompt with the matching section of the master version, in plain text. Keep the named faculty. The limits in your field are listed in Section 5. 用母版中对应的部分逐题回答,使用纯文本。保留老师姓名。本专业的篇幅要求见第 5 部分。 |
The five questions every admissions committee asks招生委员会必问的五个问题
| # | PANEL QUESTION 评审问题 | WHERE YOU ANSWER IT 在哪里回答 |
|---|---|---|
| 1 | Has this person done research, and can they describe it precisely? 申请人做过研究吗?能准确描述吗? | Research experience · CV · letters 研究经历 · 简历 · 推荐信 |
| 2 | Are the research interests clear and focused enough to start on? 研究兴趣是否清楚、聚焦到可以开始? | Research interests & questions 研究兴趣与问题 |
| 3 | Is there a faculty member here who would want to supervise this student? 这里有没有老师愿意指导这位学生? | Fit with faculty & programme 匹配 |
| 4 | Is the applicant ready for a long PhD, and do they know why they want it? 申请人是否准备好读一个长周期的博士,并清楚自己为什么要读? | Goals & readiness 目标与准备度 |
| 5 | Is the writing clear, specific and within the limit? 文字是否清楚、具体、符合篇幅? | Every paragraph · Closing & formatting 每一段 · 结尾与格式 |
The golden thread逻辑主线
A statement of purpose is an argument that you are ready to do research in this programme, so the life story can stay out. It works when each paragraph follows from the one before: an observation leads to research experience, the experience leads to a question, and the question leads to particular faculty. If you cannot write these four sentences, settle your direction before drafting anything else.
目的陈述要论证的是一件事:你已准备好在这个项目做研究,人生故事可以省掉。写得好的陈述,每一段都接着上一段:一个观察引出研究经历,经历引出研究问题,问题指向具体的老师。如果写不出下面这四句话,请先确定方向再动笔。
My dissertation model located 94% of faults within one bus, but its accuracy fell below 70% once I added rooftop solar.
I built the graph neural network in PyTorch on 1,200 simulated faults, and I wrote the data pipeline for 18 months of phasor measurements as a research assistant.
Can physics-informed graph learning locate faults reliably with the few sensors a utility can afford, as inverter-based generation grows?
Professor A’s hardware-in-the-loop lab and Dr B’s graph learning for physical systems match the testing and the modelling halves of the question.
The worked example used in this guide本指南使用的范例
Background 背景
BEng Electrical Engineering and Automation (Mainland China); MSc Electrical Power Engineering (UK), with a dissertation on graph neural networks for fault location; one year as a research assistant on a phasor-measurement project; second author of one conference paper.
Applying to 申请
PhD in Electrical Engineering at US universities: a statement of purpose of 1-2 pages naming Professor A and Dr B, with faculty also selected in the application.
Working title 暂定题目
Physics-Informed Graph Learning for Fault Detection and Location in Distribution Networks with High Inverter-Based Generation
The same statement appears in every section, so you can watch it move from its first line to its last; the whole text is printed after the sections. Do not copy the topic or the story. Study the moves and the numbered notes beside each extract, then make the same moves with your own experience, questions and faculty. Citations shown as (Authors, Year) and names such as Professor A are placeholders. The examples are in English only, because your application will be written in English.
同一份陈述贯穿每个部分,方便你看到它如何从第一句写到最后一句;各部分之后附有完整全文。不要照搬课题或经历,要学习写法:先看每段旁边的编号解析,再用你自己的经历、问题和意向导师做同样的事。(Authors, Year) 与 Professor A 等均为占位符。范例仅用英文,因为你的申请材料也用英文写。
Opening Hook开篇
Typical length 典型篇幅 60-120 words · about 10% 约占 10%The first paragraph tells the committee who you are as a researcher and what you want to study. A short, concrete moment is fine if it leads straight to your research; within three sentences the reader should know your field, your topic and the kind of work you want to do.
第一段要告诉评审:你是怎样的研究者、想研究什么。可以用一个简短具体的片段开头,但必须直接引向你的研究;三句话之内,读者应当知道你的领域、课题以及你想做的研究类型。
Blueprint结构
A specific moment, fact or result from your own work.
你自己工作中的一个具体片段、事实或结果。
What you study, in one sentence.
用一句话说明你研究什么。
The programme and the kind of work you will do.
申请的项目与你要做的研究类型。
1My MSc dissertation model, a graph neural network trained on simulated faults, located 94% of faults within one bus on a standard distribution feeder, but its accuracy fell below 70% once I added rooftop solar to the network. 2Finding out why has become my research focus: how to detect and locate faults on distribution networks where inverter-based generation makes fault currents small and unusual, using only the few measurements a utility can afford. 3I am applying to the PhD in Electrical Engineering to develop physics-informed learning methods for protection and to test them beyond simulation.
Why it works 写法解析
Opens with a result and the condition under which it broke down.
以一个结果及其失效条件开篇。
The research problem in one sentence, with its practical constraint.
一句话写出研究问题及其现实约束。
Programme, method and the step beyond simulation.
项目、方法,以及超越仿真的一步。
Draft the opening after the research experience and interests sections are finished. Then check that every claim in it is supported later in the statement.
开篇最好在研究经历与研究兴趣写完后再写,并核对开篇中的每个说法都能在后文找到支撑。
Openings that committees remember让评审记住的开篇
A childhood interest that says nothing about research ability.
童年兴趣,说明不了研究能力。
A result with a limit, which points to a research problem.
一个有局限的结果,指向研究问题。
A general claim with no problem, method or result.
泛泛的论断,没有问题、方法或结果。
A specific technical problem.
一个具体的技术问题。
The same move in other fields其他方向的同一写法
Common opening mistakes开篇常见错误
Opening with competition prizes. Put them in the CV and open with research.
以竞赛奖项开篇。奖项放进简历,开篇写研究。
Opening with a famous quotation spends words on someone else’s voice.
用名人名言开头,把篇幅花在了别人的话上。
A story that runs past 120 words pushes your research too far down the page.
超过 120 词的故事会把研究内容挤得太靠后。
Research Experience研究经历
Typical length 典型篇幅 300-450 words · about 35% 约占 35%This is the core of the statement and the part committees read most closely. For each project, say what the question was, what you did yourself, what you found and what you learned. Put the most relevant project first, not the oldest, and include a result that did not go as expected if it changed your thinking.
这是陈述的核心,也是评审读得最仔细的部分。每个项目都要写清:研究问题是什么、你本人做了什么、发现了什么、学到了什么。把最相关的项目放在最前面,而不是最早的项目;如果有结果不如预期但改变了你的思路,也应写进去。
Blueprint结构
What the study asked.
研究问的是什么。
Tasks you did yourself, with methods and tools.
你本人完成的工作、方法与工具。
What you found, including surprises.
发现了什么,包括意外结果。
How it changed your question or skills.
它如何改变了你的问题或能力。
1My MSc dissertation asked whether a graph neural network could locate faults on a distribution feeder from a small number of measurements. 2I generated 1,200 fault scenarios on the IEEE 33-bus test feeder in OpenDSS, varying fault type, location and resistance, and built a graph neural network in PyTorch that treats buses as nodes and lines as edges. I compared it with an impedance-based method on the same scenarios and wrote the evaluation code myself, reporting both the share of faults located within one bus and the mean distance error so that near misses were visible. 3The model located 94% of faults within one bus, but when I added rooftop solar at 30% of buses its accuracy fell below 70%. Inspecting the misplaced faults showed that most were high-resistance faults near buses with solar, where the inverters limited the fault current. 4That analysis taught me more than the headline result, and it is why I now want to build the physics of the network into the model rather than ask the model to learn it from data alone.
For the past year I have worked as a research assistant on a phasor-measurement project. I wrote a Python pipeline that cleans 18 months of measurement data with timestamp gaps and outliers, and I am second author of a conference paper on event detection from those data. When 11% of the records turned out to have gaps, I wrote a gap-detection step, tested three ways of filling short gaps against held-out data and documented the choice; the paper’s results were later shown to be unchanged with and without the filled data. The code is in a public repository, and the work showed me how sparse and noisy real measurements are compared with simulation. I also presented our methods to engineers at the partner utility, whose questions about false alarms changed how we reported the results.
My BEng in Electrical Engineering and Automation in Mainland China covered power-system analysis and relay protection, and my final-year project on transformer fault diagnosis was my first experience of training a classifier on measured data. It also taught me to check the physical meaning of every feature, which I still do before I train a model.
Why it works 写法解析
Opens with what the dissertation asked.
先写毕业论文研究的问题。
OpenDSS, PyTorch, 1,200 scenarios and a baseline: checkable details.
OpenDSS、PyTorch、1,200 个场景和基准方法:都是可核实的细节。
The drop in accuracy is explained, not hidden.
解释而不是掩盖准确率的下降。
The RA work adds messy measurements, a paper and public code.
研究助理经历补充了真实测量数据、论文与公开代码。
Showing what you did, not what you saw写你做了什么,而不是看到了什么
A list of tools with no task. The CV already has it.
只列工具,没有任务,简历里已经有了。
Tools named inside a task with a scale and a baseline.
在具体任务中写出工具、规模与基准。
“We”, “advanced” and “excellent” hide the applicant’s role and the numbers.
“我们”“先进”“优秀”掩盖了申请人的角色和数字。
Numbers, including the one that went wrong.
写出数字,包括出问题的那个。
Committees read the statement next to your letters of recommendation. Tell each referee which project you describe and what you did on it, so that the letter confirms the same tasks and adds what you cannot say about yourself. Most US programmes in the data ask for three letters; Duke discourages extra letters and Brown asks for at least two from faculty.
评审会把陈述与推荐信放在一起读。请告诉每位推荐人你在陈述中写了哪个项目、你在其中做了什么,让推荐信印证同样的工作,并补充你自己不便说的评价。数据中多数美国项目要求三封推荐信;杜克不鼓励多交,布朗要求至少两封来自教师。
The same move in other fields其他方向的同一写法
Common mistakes in this section本部分常见错误
Reporting accuracy without saying what it was compared with. Committees look for the baseline.
只报准确率,不说与什么相比。评审会找基准。
Listing every course and award in prose. The CV already has them.
用文字罗列所有课程和奖项,简历里已经有了。
Starting from first-year undergraduate work buries the strongest project.
从本科一年级写起,会把最重要的项目埋在后面。
Research Interests & Questions研究兴趣与问题
Typical length 典型篇幅 150-250 words · about 20% 约占 20%A statement of purpose does not need a full research design, because your project will change during the first years of the programme. It does need one or two questions that are specific enough to show you know the field, with a sentence on the methods you would use and why existing work does not answer them yet. Stay focused, and leave room for the questions to change.
目的陈述不需要完整的研究设计,因为课题会在项目的头几年发生变化。但你需要提出一到两个足够具体的问题,以表明你了解这个领域,并用一句话说明拟用方法,以及现有研究为什么还没有回答这些问题。问题要聚焦,也要给它留出变化的余地。
Blueprint结构
One or two, stated outright.
一到两个问题,直接写出。
How you would approach each.
每个问题拟用的方法。
The gap in one sentence, with a citation.
用一句话写研究空白,附引用。
How the questions might develop.
这些问题可能如何发展。
1I want to work on protection for distribution networks with high inverter-based generation. My central question is whether physics-informed graph neural networks can detect and locate faults reliably with the few sensors a utility can afford, as solar penetration and network topology change. A second question follows: where should those few sensors be placed so that the method keeps working?
2I would start on standard IEEE test feeders with varying solar penetration and then test the most promising methods in hardware-in-the-loop experiments, where real protection devices respond to simulated faults. 3Data-driven fault locators perform well in simulation, but most assume dense measurements, a fixed topology and little inverter-based generation (Authors, Year), so the conditions that utilities now face are the ones least studied. Physics-informed models have been used for state estimation (Authors, Year), but rarely for protection, where a method must work within a few cycles and must explain its decisions to the engineers who rely on it. 4I expect the methods to change as I learn more about power-electronic modelling, and the same ideas could apply to microgrids and to transmission networks with high wind generation.
Why it works 写法解析
A central question and a design question that depends on it.
一个核心问题,以及依赖其答案的设计问题。
Shows a path from the dissertation to testing beyond simulation.
展示从毕业论文到超越仿真的路径。
Three assumptions in existing work that the project removes.
现有研究的三个假设,正是本项目要去掉的。
Names where the work could go next.
写出研究的可能延伸。
Focused but open聚焦而开放
Four broad areas and no problem.
四个宽泛领域,没有具体问题。
A problem with its constraint.
一个带有约束条件的问题。
An impossible promise that suggests the applicant does not know the field.
不可能实现的承诺,显得申请人不了解这个领域。
A realistic design question.
一个现实的设计问题。
A quick test for your questions检验研究问题的小测试
| CHECK 检查项 | WHAT IT MEANS 含义 |
|---|---|
| Specific 具体 | A faculty member could name the variables, data or sources from the question. 老师能从问题中看出变量、数据或材料。 |
| Open 未解决 | You can say in one sentence why existing work does not answer it. 你能用一句话说明现有研究为什么没有回答它。 |
| Doable 可完成 | It fits a PhD at this programme with its data, labs or collections. 依托该项目的数据、实验室或馆藏能在博士期间完成。 |
| Connected 有联系 | It follows from the research experience in Section 1. 它来自第 1 部分的研究经历。 |
The same move in other fields其他方向的同一写法
Common mistakes with questions研究问题常见错误
Writing a thesis title in place of a question. State what you want to find out.
用论文题目代替问题。请写出你想弄清楚什么。
Four subfields joined by “and”. Choose the question you would start on.
用“和”串起四个子领域。请选定你会着手研究的问题。
“I will prove that…” sounds fixed before training has started.
“我将证明……”在训练开始前就已定型。
Fit with Faculty & Programme与导师和项目的匹配
Typical length 典型篇幅 150-250 words · about 18% 约占 18%Faculty often read the statements of applicants who name them, and admission can depend on one of them wanting to take you. Name two or three faculty, link each to a specific part of your question or method, and add one feature of the programme that matters for your work.
老师常常会阅读点名自己的申请人的陈述,录取也可能取决于是否有老师愿意招你。请点名两到三位老师,把每位老师与你的研究问题或方法的具体部分联系起来,并补充一项对你的研究有意义的项目特色。
Blueprint结构
The closest match, linked to your main question.
最契合的老师,对应主要问题。
A second link, often to the method.
第二位老师,常对应方法。
Training, data, facilities, centre or collection.
训练、数据、设施、研究中心或馆藏。
What the group gains from you.
课题组能从你这里得到什么。
1Professor A’s group works on protection of inverter-dominated grids (Authors, Year), and its hardware-in-the-loop laboratory would let me test methods with real protection devices rather than only in simulation. 2Dr B’s research on graph learning for physical systems (Authors, Year) matches the modelling side of my question, in particular her work on building conservation laws into graph networks, which is the step my dissertation model lacked. Working between the two groups would let me design the model with the protection constraints in mind from the start, rather than adding them after training, and to share the test cases between the two groups.
3Graduate courses in power-system dynamics, power electronics and machine learning would deepen both halves of my MSc training. 4In return, I can bring a working codebase for fault simulation and graph learning, experience of cleaning real measurement data, and a habit of releasing code that others can run. I would be glad to help maintain the group’s simulation tools and to support undergraduate projects that use them.
Why it works 写法解析
The first name is tied to a facility the project needs.
第一位老师对应项目需要的设施。
The second name is tied to the exact gap in the dissertation model.
第二位老师对应毕业论文模型的具体短板。
Training that strengthens both halves of the project.
加强项目两部分的课程训练。
Code and data skills the group can use at once.
课题组可以立即用上的代码与数据能力。
Writing fit that faculty recognise让老师认出自己的匹配段落
The committee knows its ranking; this says nothing about fit.
评审知道自己的排名,这句话与匹配无关。
A specific resource tied to the project.
与项目相连的具体资源。
The group may have moved on; the link may be out of date.
课题组可能早已转向,联系已经过时。
Shows that you checked what the group does now.
表明你核实过课题组目前在做什么。
Check that each person is still at the university and is taking students this cycle, that you have read the work you cite, and whether the programme asks you to name faculty in the form as well. MIT Chemistry asks you to name three faculty, Arizona State two or three, Northwestern Computer Science up to two, and Harvard SEAS asks you to name labs. Carnegie Mellon and Michigan let you select faculty in the application; Carnegie Mellon, the U Washington Allen School and Texas A&M say no prior contact is needed, and Brown Computer Science does not ask you to name faculty.
核实每位老师仍在该校任职且本轮招生,你引用的研究确实读过,并确认项目是否还要求在申请表中填写老师姓名。麻省理工化学要求写明三位老师,亚利桑那州立两到三位,西北大学计算机最多两位,哈佛 SEAS 要求写明实验室。卡内基梅隆与密歇根在申请表中选择老师;卡内基梅隆、华盛顿大学 Allen 学院与得州农工说明无须事先联系,布朗计算机不要求写明老师。
The same move in other fields其他方向的同一写法
Common fit mistakes匹配部分常见错误
Naming one adviser in a programme that admits by rotation. Name two or three labs and say why.
在轮转制项目中只写一位导师。请写两三个实验室并说明理由。
A paragraph copied from another application with the old school name left in.
从其他申请复制段落,忘了改学校名称。
Naming the best-known professor, who may not be taking students.
只写最知名的教授,而对方可能并不招生。
Goals & Readiness目标与准备度
Typical length 典型篇幅 80-150 words · about 10% 约占 10%Committees admit students who will finish. State a realistic goal, show the skills you already have for the first year, and, where it helps, address one gap in your record directly and briefly. A US PhD usually includes coursework and qualifying steps before the dissertation, so committees look for students who know why they want a long programme.
评审更愿意录取能够完成学业的学生。写明一个现实的目标,展示你已具备的第一年所需能力;如有必要,直接而简短地说明履历中的一个短板。美国博士通常在论文之前还有课程与资格考核,因此评审更看重清楚自己为什么要读一个长周期博士的学生。
Blueprint结构
Academic or applied, in one sentence.
学术或应用方向,一句话。
Methods and tools you use already.
你已在使用的方法与工具。
Evidence that you handle setbacks.
你能应对挫折的证据。
One sentence, facts first.
如有短板,一句话,先讲事实。
1After the PhD I plan to work on grid protection in a university or a network operator’s research team, helping utilities place the few sensors they can afford where they detect the most faults. 2I am ready to begin: I use PyTorch, OpenDSS and MATLAB every day, maintain a public repository, and have written for publication. 3A year of research assistance has also shown me that most of the work in a project lies in data and tools rather than in the model, and that a problem such as missing data is solved by testing alternatives and recording the choice. I expect hardware experiments to bring problems of the same kind, and I know how to work through them patiently.
Why it works 写法解析
Academic or industrial research, tied to the sensor question.
学术或产业研究,与传感器问题相连。
Software, code and writing the CV can confirm.
简历可以印证的软件、代码与写作。
Reuses the data-gap episode from Section 1.
沿用第 1 部分的数据缺失事件。
A claim about character with no evidence.
关于品格的自我评价,没有证据。
Persistence shown through a task.
通过一项任务展示韧性。
The same move in other fields其他方向的同一写法
Closing & Formatting结尾与格式
Typical length 典型篇幅 50-90 words · about 7% 约占 7%End by restating your research aim and why this programme is where you can pursue it, in two or three sentences. Then check the form: the limit, the spacing and whether the programme counts pages, words or characters.
结尾用两三句话重申你的研究目标,并说明为什么要在这个项目里完成它。然后检查格式:篇幅限制、行距,以及项目按页数、词数还是字符数计算。
Blueprint结构
The research aim in new words.
用新的表述重申研究目标。
The people and resources that matter.
最重要的老师与资源。
No new material, no long thanks.
不加新内容,不长篇致谢。
1I want to build fault location methods that keep working as distribution networks change, and to test them on real hardware before utilities rely on them to keep power flowing safely. 2Professor A’s laboratory and Dr B’s graph learning research make this the programme where I can do that, and I would welcome the chance to discuss the project with them.
Why it works 写法解析
The problem and the step beyond simulation, in new words.
用新的表述重申问题与超越仿真的一步。
The lab and the method, then stop.
实验室与方法,然后结束。
Format格式
| ITEM 项目 | UPLOADED DOCUMENT 上传文件 | TEXT BOX IN THE FORM 表格栏目 |
|---|---|---|
| Length 篇幅 | Use 90 to 100% of the limit. If none is given, 1 to 2 pages or about 1,000 words. 用足限制的 90% 至 100%。没有限制时,写 1 至 2 页或约 1,000 词。 | Check whether spaces count. Paste early and see where the text is cut. 确认空格是否计入。尽早粘贴,看看文字在哪里被截断。 |
| Layout 版式 | Plain font at 11 to 12 point, normal margins, the spacing the page asks for; paragraphs, not bullet points; your name and the programme at the top. 普通字体 11 至 12 号,常规页边距,按页面要求设行距;用段落而不是项目符号;页首写姓名与项目名称。 | Plain text with a blank line between paragraphs; no heading. 纯文本,段落之间空一行;不写标题。 |
| File 文件 | PDF named clearly, for example Surname_Firstname_SOP.pdf; one spelling convention throughout. PDF,命名清楚,例如 Surname_Firstname_SOP.pdf;全文使用同一种拼写习惯。 | Keep a copy of what you submitted for the interview. 保留一份提交的文本,供面试前使用。 |
Ending well, and the format好的结尾与格式
Science and engineering statements in the data are short: Stanford Electrical Engineering and Cornell Engineering up to 2 pages, Caltech about 2 pages, Carnegie Mellon and UIUC Computer Science 1-2 pages, Harvard SEAS Computer Science up to 1,000 words, UCLA Computer Science up to 500 words and Texas A&M Computer Science one page. MIT Chemistry asks for 1,000-1,500 words plus a 500-1,000-word personal statement, and UC Berkeley Mechanical Engineering 1-2 pages plus a Personal History of 1-2 pages with no overlap. UC San Diego CSE and USC Viterbi set no limit.
数据中理工科陈述都较短:斯坦福电子工程与康奈尔工程不超过 2 页,加州理工约 2 页,卡内基梅隆与伊利诺伊计算机 1-2 页,哈佛 SEAS 计算机不超过 1,000 词,UCLA 计算机不超过 500 词,得州农工计算机一页。麻省理工化学要求 1,000-1,500 词,另加 500-1,000 词个人陈述;伯克利机械工程 1-2 页,另加 1-2 页个人经历陈述且不得重复。加州大学圣地亚哥计算机与南加州大学 Viterbi 不设上限。
A large claim with nothing behind it.
空泛的大话,没有依据。
Ends on the people and the work.
以老师和研究收尾。
Breaks the format; some portals reject it.
违反格式要求,有些系统会拒收。
Meets the limit and keeps the strongest work.
既符合篇幅,又保留最强的内容。
The same move in other fields其他方向的同一写法
Common closing and formatting mistakes结尾与格式常见错误
A new project or interest in the last paragraph that the reader has not seen.
最后一段出现前文没有提过的项目或兴趣。
Uploading the personal statement in the statement-of-purpose slot, or the reverse.
把个人陈述上传到目的陈述栏,或反过来。
Ignoring a limit. Some forms cut the text and some readers stop at the limit.
忽视篇幅限制。有的表格会截断文字,有的评审读到上限就停止。
The second statement, and the one-supervisor version第二份陈述与“一位导师”版本
The Personal History or personal statement个人经历陈述或个人陈述
Many US universities ask for a second statement about you as a person: a Personal History Statement at UC Berkeley (about 1.5-2 pages), Purdue and UC Santa Barbara, a Life Experiences Statement at Duke, a separate personal statement at Cornell, and up to 500 words at Northwestern, UCLA and Michigan. The prompt differs by university, so answer the one on the page. Keep the two apart: the statement of purpose is about your research; the second statement is about your path and what you would add to the department. At Brown and Boston University the personal statement is the main essay and no separate statement of purpose is listed; USC also calls its statement a personal statement. Write it as a statement of purpose, with research experience, questions and fit, and keep one short paragraph on your path.
许多美国大学还要求一份关于你个人的陈述:加州大学伯克利(约 1.5-2 页)、普渡与加州大学圣塔芭芭拉要求个人经历陈述,杜克要求人生经历陈述,康奈尔要求单独的个人陈述,西北大学、UCLA 与密歇根不超过 500 词。各校题目不同,请按页面题目回答。两份文书分开写:目的陈述写研究;第二份陈述写你的成长路径,以及你能为院系带来什么。布朗大学与波士顿大学以个人陈述为主文书,未单列目的陈述;南加州大学也把陈述称为个人陈述。此时请按目的陈述的结构来写(研究经历、问题与匹配),只用一小段写个人经历。
1I grew up in a town where power cuts were common in summer, and my father worked as an electrician for the local grid company. 2At university I led a volunteer team that repaired lighting and wiring in rural primary schools, visiting 12 schools over two summers.
3The work taught me to explain technical problems to people without technical training, and to plan around missing tools and parts. 4As a PhD student I would like to help with outreach to school students and to mentor undergraduates who are new to research.
Why it works 写法解析
Background in one sentence, linked to the field.
一句话交代背景,并与领域相连。
A team the applicant led, with a number.
申请人带领的团队,附数字。
Skills the statement of purpose has no room for.
目的陈述中放不下的能力。
Outreach and mentoring the department can use.
院系能用上的外展与指导。
When one supervisor must agree first须先取得一位导师同意时
Most US universities in the data say there is no need to contact faculty before applying (MIT, Stanford, Michigan, NYU). A few programmes admit you to one faculty member: Wisconsin Psychology admits by individual faculty, Arizona State C&I Language and Literacy needs an approved mentor who agrees to advise you, and UC Davis Psychology strongly recommends emailing potential mentors. For these, write to the person first and narrow the statement to one project and one supervisor, as below.
数据中多数美国大学说明申请前无须联系老师(麻省理工、斯坦福、密歇根、纽约大学)。少数项目按导师录取:威斯康星心理学由各位老师单独录取,亚利桑那州立 C&I 语言与读写方向须有同意指导的导师,加州大学戴维斯心理学强烈建议给潜在导师发邮件。申请这类项目时,先联系导师,再把陈述聚焦到一个项目和一位导师,写法如下。
1I am applying to the PhD in Electrical and Computer Engineering to work with Professor A on fault location in distribution networks with high solar generation. 2I wrote to Professor A in October with my dissertation results, and we discussed how her hardware-in-the-loop laboratory could test a physics-informed version of my model. 3My proposed project would develop that model on standard test feeders and then validate it on the laboratory’s real-time simulator.
Why it works 写法解析
Names the supervisor and the problem in the first sentence.
第一句就写明导师和问题。
Says what was discussed, without claiming a promise.
说明讨论内容,不声称获得承诺。
One method and one facility.
一种方法、一项设施。
The full statement, start to finish完整陈述范例
Here is the worked example as one document of 1,001 words: the master version you cut from. Read it once without the notes to hear how the parts join, then check that each paragraph does one job. Do not reuse its sentences; use it to check proportions. For a one-page version, keep the opening, the dissertation paragraph, the central question and the first fit paragraph.
下面把范例合成一份完整文件,共 1,001 词,即你裁剪各版本时使用的母版。先不看解析通读一遍,感受各部分如何衔接,再检查每一段是否只承担一项任务。不要套用其中的句子,用它来检查各部分的比例。如果要压缩到一页,保留开篇、毕业论文一段、核心问题和第一段匹配内容。
Worked example · Statement of purpose, PhD in Electrical Engineering范例 · 完整陈述
My MSc dissertation model, a graph neural network trained on simulated faults, located 94% of faults within one bus on a standard distribution feeder, but its accuracy fell below 70% once I added rooftop solar to the network. Finding out why has become my research focus: how to detect and locate faults on distribution networks where inverter-based generation makes fault currents small and unusual, using only the few measurements a utility can afford. I am applying to the PhD in Electrical Engineering to develop physics-informed learning methods for protection and to test them beyond simulation.
My MSc dissertation asked whether a graph neural network could locate faults on a distribution feeder from a small number of measurements. I generated 1,200 fault scenarios on the IEEE 33-bus test feeder in OpenDSS, varying fault type, location and resistance, and built a graph neural network in PyTorch that treats buses as nodes and lines as edges. I compared it with an impedance-based method on the same scenarios and wrote the evaluation code myself, reporting both the share of faults located within one bus and the mean distance error so that near misses were visible. The model located 94% of faults within one bus, but when I added rooftop solar at 30% of buses its accuracy fell below 70%. Inspecting the misplaced faults showed that most were high-resistance faults near buses with solar, where the inverters limited the fault current. That analysis taught me more than the headline result, and it is why I now want to build the physics of the network into the model rather than ask the model to learn it from data alone.
For the past year I have worked as a research assistant on a phasor-measurement project. I wrote a Python pipeline that cleans 18 months of measurement data with timestamp gaps and outliers, and I am second author of a conference paper on event detection from those data. When 11% of the records turned out to have gaps, I wrote a gap-detection step, tested three ways of filling short gaps against held-out data and documented the choice; the paper’s results were later shown to be unchanged with and without the filled data. The code is in a public repository, and the work showed me how sparse and noisy real measurements are compared with simulation. I also presented our methods to engineers at the partner utility, whose questions about false alarms changed how we reported the results.
My BEng in Electrical Engineering and Automation in Mainland China covered power-system analysis and relay protection, and my final-year project on transformer fault diagnosis was my first experience of training a classifier on measured data. It also taught me to check the physical meaning of every feature, which I still do before I train a model.
I want to work on protection for distribution networks with high inverter-based generation. My central question is whether physics-informed graph neural networks can detect and locate faults reliably with the few sensors a utility can afford, as solar penetration and network topology change. A second question follows: where should those few sensors be placed so that the method keeps working?
I would start on standard IEEE test feeders with varying solar penetration and then test the most promising methods in hardware-in-the-loop experiments, where real protection devices respond to simulated faults. Data-driven fault locators perform well in simulation, but most assume dense measurements, a fixed topology and little inverter-based generation (Authors, Year), so the conditions that utilities now face are the ones least studied. Physics-informed models have been used for state estimation (Authors, Year), but rarely for protection, where a method must work within a few cycles and must explain its decisions to the engineers who rely on it. I expect the methods to change as I learn more about power-electronic modelling, and the same ideas could apply to microgrids and to transmission networks with high wind generation.
Professor A’s group works on protection of inverter-dominated grids (Authors, Year), and its hardware-in-the-loop laboratory would let me test methods with real protection devices rather than only in simulation. Dr B’s research on graph learning for physical systems (Authors, Year) matches the modelling side of my question, in particular her work on building conservation laws into graph networks, which is the step my dissertation model lacked. Working between the two groups would let me design the model with the protection constraints in mind from the start, rather than adding them after training, and to share the test cases between the two groups.
Graduate courses in power-system dynamics, power electronics and machine learning would deepen both halves of my MSc training. In return, I can bring a working codebase for fault simulation and graph learning, experience of cleaning real measurement data, and a habit of releasing code that others can run. I would be glad to help maintain the group’s simulation tools and to support undergraduate projects that use them.
After the PhD I plan to work on grid protection in a university or a network operator’s research team, helping utilities place the few sensors they can afford where they detect the most faults. I am ready to begin: I use PyTorch, OpenDSS and MATLAB every day, maintain a public repository, and have written for publication. A year of research assistance has also shown me that most of the work in a project lies in data and tools rather than in the model, and that a problem such as missing data is solved by testing alternatives and recording the choice. I expect hardware experiments to bring problems of the same kind, and I know how to work through them patiently.
I want to build fault location methods that keep working as distribution networks change, and to test them on real hardware before utilities rely on them to keep power flowing safely. Professor A’s laboratory and Dr B’s graph learning research make this the programme where I can do that, and I would welcome the chance to discuss the project with them.
Cutting the master version如何裁剪母版
| SECTION 部分 | JOB 任务 | ABOUT 500 WORDS 约 500 词 | CHARACTER LIMIT OR SET QUESTIONS 字符限制或固定问题 |
|---|---|---|---|
| 0 Opening 开篇 | Field, topic, kind of work 领域、课题、研究类型 | One sentence 一句话 | First line of the first answer 第一个回答的首句 |
| 1 Research experience 研究经历 | Proof you can do research 证明研究能力 | Strongest project only 只留最强的项目 | The “experience” prompt 对应“经历”类问题 |
| 2 Interests & questions 研究兴趣与问题 | Where you want to go 想往哪里走 | First question and one line of method 第一个问题加一句方法 | The “interests” prompt 对应“兴趣”类问题 |
| 3 Fit 匹配 | Who would supervise you 谁会指导你 | Two names, one link each 两位老师,各一处联系 | The faculty field or prompt 对应导师栏或问题 |
| 4 Goals & readiness 目标与准备度 | Why you will finish 为何能完成 | One sentence 一句话 | The “goals” prompt 对应“目标”类问题 |
| 5 Closing 结尾 | The case in brief 简要总结 | Merge into fit 并入匹配部分 | Leave out 不写 |
Adapting to each university按校调整
Keep the master version and cut from it. Never write a new statement for each programme: change the fit section, answer the programme’s own prompt, and trim to its limit, in the way the row for each target describes.
保留母版并在此基础上裁剪。不要为每个项目重写:调整匹配部分,回应项目自己的题目要求,并按下表各行的做法裁到规定篇幅。
| TARGET 目标 | WHAT TO DO 做法 |
|---|---|
| About 500 words or one page 约 500 词或一页 UCLA Computer Science (up to 500 words); Texas A&M CSE (one page) UCLA 计算机(不超过 500 词);得州农工计算机(一页) | Keep one sentence of opening, your strongest project, one question and two named faculty. Drop the undergraduate project and cut the goals to one sentence. 保留一句开篇、最强的一个项目、一个问题与两位老师。删去本科项目,目标压缩为一句。 |
| About 1,000 words or up to two pages 约 1,000 词或不超过两页 Stanford EE (2 pages single-spaced); Carnegie Mellon CS, UIUC (1-2 pages); Harvard SEAS (1,000 words) 斯坦福电子工程(单倍行距 2 页);卡内基梅隆计算机、伊利诺伊(1-2 页);哈佛 SEAS(1,000 词) | Use the master version. Rewrite the fit section for each programme and check whether the limit is in words or pages, and whether pages are single- or double-spaced. 使用母版。针对每个项目改写匹配部分,并确认限制按词数还是页数计算,以及单倍还是双倍行距。 |
| Three pages or more 三页或更长 MIT Chemistry (1,000-1,500 words naming 3 faculty) 麻省理工化学(1,000-1,500 词,写明 3 位老师) | Add detail to the research interests: methods, data or sources and a first study. Do not pad the opening. 在研究兴趣部分补充方法、数据或材料与第一项研究。不要拉长开篇。 |
| A character limit, a form or set questions 字符限制、表格或固定问题 Penn Engineering (six prompts in 2 pages); Georgia Tech BME (5,000 characters) 宾大工程学院(2 页内回答六个问题);佐治亚理工生物医学工程(5,000 字符) | Answer each prompt with the matching section of the master version, in plain text, and paste the draft early to see where it is cut. Keep the named faculty. 用母版中对应的部分逐题回答,使用纯文本,并尽早粘贴草稿,看看在哪里被截断。保留老师姓名。 |
| Statement of purpose plus a personal statement 目的陈述加个人陈述 UC Berkeley ME (Personal History, no overlap); Cornell; UCLA; Purdue; UC Santa Barbara 伯克利机械工程(个人经历陈述,不得重复);康奈尔;UCLA;普渡;加州大学圣塔芭芭拉 | Move personal background out of the statement of purpose into the second statement, so the two do not overlap. 把个人背景从目的陈述移到第二份陈述,避免两者重复。 |
The email before the application申请之前的那封邮件
Where a programme encourages contact (Carnegie Mellon, Georgia Tech and UC Davis Psychology), your first email is part of the application: give enough detail for the faculty member to judge the fit, with one piece of their work, your question, what you have done towards it and your CV. Where a programme asks you not to contact faculty (Cornell English before an offer) or says no contact is needed (the U Washington Allen School), name people in the statement instead. Write the first draft yourself; your academic will check it.
在鼓励联系老师的项目(卡内基梅隆、佐治亚理工与加州大学戴维斯心理学),第一封邮件就是申请的一部分:写得足够详细,让老师能判断是否匹配,包括对方的一项研究、你的研究问题、你已做过的相关工作,并附简历。要求不要联系(康奈尔英文系录取前)或说明无须联系(华盛顿大学 Allen 学院)的项目,只在陈述中写明老师即可。第一稿由你自己写,导师负责核对。
Read it like a panel用评审的眼光重读
Reread your statement as a faculty member with a few minutes and a long list of files. Score each question 1 (weak), 2 (adequate) or 3 (strong), and fix the lowest score first.
像一位只有几分钟、还有一长串材料要读的老师那样重读。每项打 1(弱)、2(一般)、3(强)分,先修改得分最低的一项。
| # | CHECK 检查 | SELF-SCORE 自评 |
|---|---|---|
| 1 | Does the first paragraph name my field and topic within three sentences? 第一段三句话内是否写明了领域与课题? | 1 2 3 |
| 2 | For each project, can the reader see the problem, my own tasks, the tools, the baseline and the result? 每个项目是否都写清了问题、我本人的工作、工具、基准和结果? | 1 2 3 |
| 3 | Are my research questions specific, open and connected to my experience? 研究问题是否具体、尚未解决,并与经历相连? | 1 2 3 |
| 4 | Is every named faculty member linked to a question or method, and taking students? 每位被点名的老师是否都对应问题或方法,且本轮招生? | 1 2 3 |
| 5 | Do the goals and skills show I can finish a long programme? 目标与技能能否表明我能完成一个长周期的项目? | 1 2 3 |
| 6 | Is personal background in the right document, with no overlap between the two statements? 个人背景是否放在正确的文书里,两份陈述是否没有重复? | 1 2 3 |
| 7 | Is it within the limit and format for this programme, with the right school name throughout? 是否符合该项目的篇幅与格式,学校名称是否全部正确? | 1 2 3 |
Three tests三个测试
Would the professors you name recognise their own work in your fit section? If not, rewrite it.
被点名的教授能在你的匹配段落中认出自己的研究吗?不能就重写。
Replace the school name with another. If the paragraph still works, it is not specific enough.
把学校名换成另一所。如果段落仍然成立,说明写得不够具体。
Explain your research experience and your question in two minutes. Where you stumble, the writing is unclear.
用两分钟讲清你的研究经历和研究问题。卡壳的地方就是写得不清楚的地方。