Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Ai Carrer Guide

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

ARTIFICIAL

INTELLIGENCE
CAREER GUIDE
Student Affairs at Carnegie Mellon University Silicon Valley
2

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

CONTENTS

INTRODUCTION......................................................................... 3

EXAMPLE RESUME.......................................................................4

WHAT IS A CAREER IN AI?...........................................................5

COURSEWORK AND PROJECTS.................................. . . . . . . . . . ........6

RESEARCH EXPERIENCE............................................................9

P A T H W A Y S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

ARTIFICIAL INTELLIGENCE
CAREER GUIDE

Acknowledgment section Introduction

The purpose of this guide is to help you understand what a career in Artificial Intelligence
These career guides are adapted each (AI) means, and how to prepare for a job search in various areas of AI. Example roles
year, and the career services team within artificial intelligence include but not limited to the roles of Data Scientist,
would like to sincerely thank the 2019
cohort of Peer Career Consultants for
Machine Learning Engineer, Deep Learning Engineer, and Machine Learning
creating the first iteration. We appreciate Software Engineer. The materials herein are intended as a starting point for students
the hard work of Chaitra Dinesh Pai, to use and prompt further research into strategic planning for a career in the field of AI.
Nursultan Dyussebayev, Arpit Gupta,
Sweta Hari Kumar, and Joseph Konan
As the #1 leading school in the field of Artificial Intelligence according to U.S. News and
who researched specific fields of interest World Report, Carnegie Mellon University is consistently pushing the bounds of what is
and compiled helpful tips and resources
considered state-of-the-art with world-leading faculty, prestigious research labs, and a
for students to enable them to make the
most of their CMU-SV experience and diverse network of interdisciplinary researchers, alumni, and industry sponsors.
navigate the internship and job search
process in Silicon Valley. Their valuable
contributions made these career guides
possible. Additionally, the example
resumes are compiled from recent
alumni who also wanted to help our
future students at CMU-SV. Thank you
to everyone who contributed their time
and effort.

Image: Carnegie Mellon Artificial Intelligence Beats Chinese Poker Players by $792,327
4

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

EXAMPLE To give you an idea of what an artificial intelligence focused resume can look like before it’s
tailored to the job description, please see the example below. *Please note, these roles may not

RESUME
have occurred in 2019-2020. This is a resume example from a real CMU-SV student, however,
other data points are edited each year. Please recognize that this is not a real candidate. Please
email career-services@sv.cmu.edu if you have any questions about this example resume or if
you need to schedule a resume appointment, please do so through Handshake.

Mountain View, CA
linkedin.com/in/cmutartan1 CM U T AR T AN (555) 555-5555
Cmutartan1@andrew.cmu.edu

EXPERIENCE
Graduate Research Assistant Emirates Airlines May 2019 – Present
CMU-Emirates iLab
• Reduced travel time variability of intercontinental flights by 75% by implementing LSTM TensorFlow algorithm.

• Implemented iLab integration with OS X Spotlight Search by creating tool which extracts metadata and uses natural language

processing on saved chat transcripts to provide a 148% faster system-wide search database.
• Redesigned chat data pipeline and implemented cloud compatibility for search, scaling to 30,000+ users.

Graduate Research Assistant Carnegie Mellon University January 2019 – May 2019
Accountable Systems Lab
• Identified high influence neurons in quantity and distribution of interest, using influence-directed measures.

• isolated individual features the network uses to make decisions and distinguish related classes using PyTorch.

Graduate Teaching Assistant Carnegie Mellon University August 2018– Present


• Courses: Machine Learning for Product Managers, Intro to Deep Learning, Intro to Machine Learning.

• Mentoring 20+ students on creating state-of-the-art AI. Also, product owner of 4 end-to-end data-driven apps.

EDUCATION
Mountain View, CA Carnegie Mellon University Aug 2018 – May 2020
• M.S. in Electrical and Computer Engineering

• Graduate Coursework: Machine Learning for Signal Processing; Hardware Architectures for Machine Learning; Speech
Recognition and Understanding; Machine Learning for Product Managers; Intro to Deep Learning.

Made-in Abyss, Japan Orth Prefecture University Aug 2014 – May 2018
• B.S. in Computer Science and Engineering with Minor in Mathematics

• Undergraduate Coursework: Artificial Intelligence; Mathematical Statistics; Advanced Calculus; Advanced Linear Algebra;
Operating Systems; Databases; Algorithms; Programming Languages; Computer Architecture.

TECHNICAL PROJECTS

• Applications of Deep Learning in Activity Recognition (2018). Python, PyTorch, TensorFlow, ONNX. Center Loss & Wav2Letter,
classifying 9 activity vibrations. 80% raw signal accuracy. 90% feature engineering accuracy.
• Self-Attention Networks for Connectionist Temporal Classification (2002). Python, PyTorch. Speech recognition for transcript
generation using Bi-LSTM recurrent neural net with string edit difference of 16.8.
• Deep Residual Learning for Image Classification and Verification (2003 – 2004). Python, PyTorch, OpenCV.
Leveraged CNN residual networks, augmentation, alignment. 69% classification accuracy. 0.93 verification AUC.
• Sparse Multilayer Perceptron for Phoneme Recognition (2004). Python, PyTorch. Created high-efficiency low-parameter
network to identify phoneme state labels using raw mel spectrogram frames. 65% accuracy.
• Deep Learning Automatic Differentiation Libraries (2002). Python. Developed functions and methods from scratch for MLP,
CNN, and RNN forward/backward-propagation, gradient descent, and attention mechanism.

WORKING PAPERS AND PUBLICATIONS


• NIPS 2018. Generalized Inverse Optimization through Online Learning. Nanachi Hollow, Riko Delver, Reg Abyss.

• IEOM 2017. Neural Networks Applications in Manufacturing Processes. Mitty Narehate , Nanachi Hollow.

Languages and Technologies


• Python, R, SQL, C/C++ | SciKit-Learn, TensorFlow, PyTorch, Keras, ONNX, Matplotlib, ggplot2, Seaborn, Plotly

• Jupyter, RStudio, PyCharm, Ubuntu, Linux | AWS, EC2, SageMaker, Google Cloud, CUDA
5

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

CAREER GUIDE STARTER KIT

This guide is broken into five parts, which aim to outline: learning, getting an internship or full-time role working on
career expectations, academic preparation, career strategies, artificial intelligence is significantly more difficult. Employers
presentation, and a case-study of how to make the most of like to see relevant, high impact projects that have been
your time at CMU. deployed, published, or have won Kaggle competitions.

What Does a Career in AI Look Like? Students who are serious about getting involved in this field
are expected to take Introduction to Deep Learning 11-785.
The field of artificial intelligence uses machine learning to Successful students gain access to the network of other
directly help companies realize higher profits or to provide students who have completed the course -- connections
additional features to a product. Practitioners are able to directly relevant to the field and industry. The course is
predict things in reality by having a machine learn parameters well rounded in terms of concepts. It helps us understand
that give the best results for a provided mathematical model. the fundamentals of Deep Learning. The course starts
Example applications include predicting ad revenue in any off gradually with MLPs and progresses towards more
company, face or voice recognition on iPhones, language complicated concepts such as attention and sequence-to-
translation on YouTube, medical diagnostics, and more. sequence models. We get a complete hands on experience
Most practitioners specialize in one or more of the following: with PyTorch which is very important to implement Deep
computer vision, natural language processing, signal Learning models. As a student, you will learn the tools
processing, financial modeling, recommendation system, required for building Deep Learning models. The homework
robotics, and/or theory. assignments usually have 2 components which is Autolab
and Kaggle. The Kaggle components allow us to explore
How Do I Prepare for a Career in AI? multiple architectures and understand how to fine-tune
and continuously improve models. The tasks for all the
The most common roles in artificial intelligence involve data assignments were similar and it was interesting to learn how
science, machine learning, and more recently the subfield of the same task can be solved using multiple Deep Learning
deep learning. Of the three, deep learning is one of the most approaches. Overall, at the end of this course you will be
important and often required skills that companies leveraging confident enough to build and tune Deep Learning models.
artificial intelligence are looking for. If you don’t know deep
6

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

COURSEWORK AND PROJECTS

Having a strong project portfolio is vital to having a strong


application. At Carnegie Mellon, you have access to world
class faculty who will help you recreate state-of-the-art
results, design scalable infrastructure, or implement complex
algorithms. Relevant courses include:

Group 1: Foundations (1 Required)


18661 Intro to Machine Learning for Engineers
11785 Introduction to Deep Learning (Recommended)

Group 2: Working with data (1 Required)


18656 Data-Intensive Workflow Development
14736 Distributed Systems
14848 Cloud Infrastructure
15619 Cloud Computing

Group 3: Specialization 1 (1 Required)


18781 Speech Recognition and Understanding
18793 Image and Video Processing
18797 Machine Learning for Signal Processing
11411 Natural Language Processing

Group 4: Specialization 2
18663 Hardware Architectures in Machine Learning
18739 Special Topics in Security: Security and Fairness of
Deep Learning
49795 Artificial Intelligence for Product Managers

Group 5: Research Experience (2 Required)


18980 M.S. Graduate Project (repeatable)
NA Course Project with Advisor
NA AI Internship

Group 6: Other
18645 How to Write Fast Code
18751 Applied Stochastic Processes
7

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

COURSEWORK AND PROJECTS

Below are concrete examples of course projects as they would appear on your resume.

Self-Attention Networks for Connectionist Temporal Classification


• Developed end-to-end speech recognition encoder-decoder RNN. Generated transcripts, surpassing human-level
performance w/ Levenshtein distance of XX.X . PyTorch, Python.

Deep Residual Learning for Image Classification and Verification


• Developed deep residual learning CNN architecture for face identification with XX.X% accuracy and 0.XX AUC, leveraging
center loss and triplet loss. PyTorch, Python.

Sparse Multilayer Perceptron for Phoneme Recognition


• Created high-efficiency low-parameter MLP network to identify phoneme state labels using raw mel spectrogram frames.
XX.X% accuracy. PyTorch Python

Deep Learning Automatic Differentiation Libraries


• Implement low-level logic of an end-to-end deep learning framework, together with automatic differentiation and
optimization or MLPs, CNNs, and RNNs. PyTorch, Python.

Eigenfaces for Face Recognition and Bounding


• Implemented eigenfaces as an unsupervised dimensionality reduction algorithm, leveraging results by reshaping
eigenvectors for principal component analysis. Matlab.

Source Separation w/ Single Channel Independent Component Analysis


• Implemented my own ICA based on freeing fouth moments method, extracting independent components to simultaneously
speaking sources from a recording. Matlab.

Signal Separation using Non Negative Matrix Factorization


• Separated mixed signal into component signals, extracting speech from recordings with background sound and music by
minimizing KL divergence between spectrograms. Matlab.

Sparse Recovery of Streaming Signals Using Iterative Hard Thresholding


• Implemented iterative hard thresholding with projected gradient descent algorithm, recovering sparse image transform from
3 compressed measurements. Python, Matlab.

Influence-Directed Explanations for Deep Convolutional Networks


• Developed tool explaining neural networks by identifying influential concepts generalized across instances and isolated
individual features uses for decisions. TensorFlow, Python.

Multi-Targeted Adversarial Evasion Attack of Deep Neural Networks


• Implemented multi-targeted adversarial example that attacks multiple models within each target class with single modified
image with 100% attack success. TensorFlow, Python.
8

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

Below are concrete examples of course projects as they would appear on your resume.

Model Memorization for Calibrated White-Box Membership Inference


• Implemented Membership Inference (MI) attacks, exploiting a learned model’s lack of generality to infer whether given
sample was in a model’s training set. TensorFlow, Python.

Speech Differentiation via Naive Bayes for Text Classification


• Developed a classifier that distinguishes speeches given by Republicans (Red) and Democrats (Blue) running for president.
XX.X% accuracy. Python.

Translation via Part-of-speech Tagging and Hidden Markov Models


• Generated and evaluated first order bigram HMMs by leveraging Viterbi algorithm to calculate transmission and emission
probabilities, deriving parts-of-speech. Python.

Deep Learning for Sentiment Analysis


• Developed deep neural network to understanding feedback in product reviews and user’s opinions in tweets, identifying
positive, negative, or neutral text. PyTorch, Python.

Big Data Twitter Analytics Web Service Design and Implementation


• Developed data warehouse complete with front-end query response system, backend database, ETL tool, to process
terabytes of tweets on AWS. Java, Tomcat, Python, Hbase.

Blockchain Ledger Design and Implementation


• Developed blockchain mining application with ledger that will award coins if a block is successfully solved given a difficult
work and add a new block into the network. Go, Java.

Distributed File System Design and Implementation


• Implement a simple distributed file system using my own RMI library, allowing files to be hosted remotely on one or more
storage servers complete with API. Go, Java.

Distributed Key-value Store Design and Implementation Using Raft


• Implement Raft algorithm as a replicated state machine protocol, achieving fault tolerance by storing copies of data on
multiple replica servers. Go, Java.

Bigram ETL Tool Design and Implementation Using Apache Spark


• Counts number and frequency of bigrams, as well as the number of bigrams required to add up to 10% of all bigrams within
corpus. Apache Spark, Java.

BigTable, a Scalable NoSQL Database Design and Implementation


• Implemented a distributed storage system for managing structured data, designed to scale to a very large size: petabytes of
data across thousands of commodity servers. Java.

Multi-path TCP Analytic Tool of Instant Virtual Networks in Mininet


• Analyzed Three-Tier and Fat Tree network topologies as well as east-west congestion at the transport level using MPTCP
approach by simulating network traffic. Mininet, Python.

Remote Method Invocation API Tool for Distributed Microsystems


• Implemented remote method invocation (RMI) library for calls over network connections, permitting objects located on one
Java virtual machine to call methods on another. Java.
9

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

RESEARCH EXPERIENCE

Students who are interested in careers in AI are expected to • Movie Review Web Data: Amazon Prime Video (https://www.
have done research with a faculty member that specializes in kaggle.com/dm4006/amazon-movie-reviews)
the domain. In particular, research assistantships are easily • Stock Price Data: AMZN (Amazon) (https://finance.yahoo.
attainable and lead to high impact projects given you are com/quote/AMZN/history/)
working with faculty members at Carnegie Mellon University.
For perspective, a Research Scientist in industry will have Applying to Amazon and having multiple projects on Amazon
hundreds of citations. In contrast, ECE and SCS professors will most likely get your application noticed by Amazon.
will have thousands of citations. High-tier internships have Moreover, a code you write for these specific datasets
“a publication in top journals” as a sought after or required relevant to Amazon can be ported to get similar insights and
qualification, and you will be more successful if you publish in results for other companies. To emphasize this point, here
your interest area. are some equivalent Google datasets for those listed above:

Many courses, including 11-785, 14-736, 14-848, 18-781, 18-


• Google Play Store App Data
793, and 18-797, require you to create a project proposal and
realize that idea by getting results. These courses, in addition (https://www.kaggle.com/lava18/google-play-store-apps)
to others not listed, allow you to have a faculty advisor on • TensorFlow Speech Recognition Challenge: Google Home
your project. This means, you are able to choose a research (https://www.kaggle.com/c/tensorflow-speech-recognition-
idea that the faculty member is interested in working on challenge/data)
and get research experience while doing your coursework. • Google Analytics Customer Revenue Prediction (https://
In addition, 18-980 is an option for students who have a www.kaggle.com/c/ga-customer-revenue-prediction/data)
research project not linked to a traditional course, allowing
• Video Comments and Metadata: YouTube (https://
you to get core credit for doing research with faculty. There is
developers.google.com/youtube/v3/docs/comments)
no excuse for not having at least one research assistantship
• Stock Price Data: GOOG (Google) (https://finance.yahoo.
working with faculty that specialize in your area of interest.
com/quote/GOOG/history/)
Further, if you are starting at CMU with no relevant
experience in the field, it is readily attainable.
And more datasets for Amazon (https://www.kaggle.com/
datasets?search=Amazon) or datasets for Google (https://
If you want to work at a specific company doing AI research
www.kaggle.com/datasets?search=Google) are out there, and
and development, then cut to the chase. People who want to
you can simply search for your top company’s data so you
work at Amazon should have research projects improving the
can make the best resume for your application. Joining Kaggle
products or experience of Amazon’s engineers or customers.
competition is also a great way to build your resume: (https://
A simple google search yields multiple big-data resources that
www.kaggle.com/competitions).
are used by Amazon and researchers:

Another application of ML is Computer Vision, which is widely


• SNAP Web Data: Amazon Reviews (https://snap.stanford. used in the self driving and robotics industries.
edu/data/web-Amazon.html) Popular datasets: Waymo One (https://waymo.com/open/
• Topical Chat Dataset: Amazon Alexa (https://github.com/ data/perception/). They also hold self-driving specific
alexa/Topical-Chat) competitions (https://waymo.com/open/challenges/)
• Sales Rank Data: Amazon Print and Kindle Books (https://
www.kaggle.com/ucffool/amazon-sales-rank-data-for-print-
and-kindle-books)
10

A R T IF ICIAL INTELLIG ENCE CAREER GUI DE


S T U D ENT AFFAIRS AT CARNEGIE MELLON UNIVERSITY SILICON V A L L EY

PATHWAYS

Depending on the amount of AI related coursework you take, DS and ML Roles at Mid-tier Companies. These roles are
and the extent to which you engage with faculty, it may make not hard to find because Data science and AI is so popular.
sense for you to consider a variety of options. There are many people who want to break into data science,
and they tend not to first consider bootcamps or fellowships
Data Science Bootcamps. These, in terms of (though they should) so the applicant pool is very large. To
competitiveness, are among the least competitive given you stand out at mid-tier companies, it is not necessary that
are an engineering student from Carnegie Mellon University. you have published, but you should have relevant project
Bootcamps often cost money, so, like school, you are paying experience on the same problem they are working on.
to build your human capital. Even better, if you got results using their data or improved
their product, then you will be taken more seriously by the
Data Science Fellowships. These are like Data Science company you are applying to. The prestige of this is between
Bootcamps but are more competitive and are often Fellowships and AI Residencies, and pays similar to the latter.
sponsored by employers. These opportunities are free, and,
for this reason, there is a more competitive process getting Consider these websites if you want to find mid-tier Data
in. People who did their Ph.D. in Physics or some non-related Science Jobs. (https://www.springboard.com/blog/data-
STEM field really like these options, because they are often science/data-science-jobs/)
overqualified in terms of foundational understanding. Being
a student who has more relevant coursework will give you AI Roles at High-tier Companies. These are competitive.
more intuition than the people who are very smart but want a You should have: an AI residency on a related subject, an
career change. internship (think short residency) with the company and
published, won a major competition by the company and
AI Residencies. These are ideal if you are shooting for a published, a Ph.D. dissertation on improving their product,
competitive AI role, but have either published in a low- to or equivalent. Talk to your advisors and mentors who make
mid-tier journal or have not published at all. Even if you these possible.
have published in a high-tier journal, these are still worth
considering because you are able to do similarly impactful
research but in an industry setting. AI Residencies are
full-time roles that pay like full time roles. This is becoming
more favorable among recent Ph.D. grads who do not want
to do a Post-Doc that pays subsistently while still getting
more specialized research experience. For this reason, AI
Residencies, though open to B.S. to Ph.D. students, are often
more difficult to get into than Data Science Fellowships.

Consider this list if you are interested in AI Residencies.


(https://github.com/dangkhoasdc/awesome-ai-residency)

You might also like