SRI-summer-internship

SRI-summer-internship

SRI International: Research Internships in Artificial Intelligence
Summer 2018

The Artificial Intelligence Center at SRI International is hiring for the research intern positions described
below. Positions are available in both Menlo Park, CA, and San Diego, CA. Interested parties should
contact Nick Marozick (nick.marozick@sri.com) for further details.


Positions in Menlo Park, CA
Explainable Autonomy. With the powerful but opaque nature of today’s AI algorithms, interpretability
has become a hot topic. The Explainable Autonomy project tackles the problem of autonomous agents
knowing when, what, and how to explain their behavior to human collaborators. We are looking for a
summer intern with a keen interest in collaborative AI to develop ideas and implement technology for
explainable autonomy. You will be working with experts in autonomy, machine learning, and humancomputer
interaction to design and build explainable systems. Programming experience and
knowledge of machine learning and/or autonomous agents a must; familiarity with reinforcement
learning a plus.
Proactive Decision Support. Intelligent assistant technology today is primarily user-driven, with the
system simply responding to explicit user commands. The Proactive Decision Support project aims to
develop capabilities to enable an intelligent system to take the initiative and provide timely, contextrelevant
assistance to users. We are looking for a summer intern with innovative ideas and excellent
programming skills to develop and implement algorithms for proactive decision support. Prior and
planned research threads for the project include sequential pattern mining, plan recognition, and text
analytics. Programming experience and knowledge of machine learning algorithms a must; familiarity
with text analytics a plus.
Whole-body Person Recognition. This project seeks to recognize people from whole-body videos,
exploiting aspects of body shape and motion. The research will include both the development and
evaluation of a state-of-art identification system on existing data collections, as well as a more
systematic exploration of the effects of difficult viewing conditions on whole-body identification by
using rendered 3D motion capture data. Our objectives are both to push the state of the art as well as
to understand the nature of the problem as an aid to future research programs.
Detection of Tampered Videos. The goal is to detect tampering of videos, such as the replacement of a
segment of audio or the replacement of a sequence of video frames. The research will focus on
techniques that combine extracted descriptions of both the audio and visual components of a video,
pinpointing discrepancies between them. For example, and audio analysis indicating that the audio
was recorded in a small room (by examining reverberation properties of the audio track) combined
with a visual analysis indicating that the scene is outdoors could indicate tampering.


Positions in San Diego, CA
Friendly Bot. The aim of this project is to develop a natural language dialog capability that
emulates the response of typical native speakers of American English to communications from
individuals unknown to them. The intern will study available communication corpora and develop a
modest domain theory that encompasses a handful of intents and language acts relevant to a bot
responding in a friendly fashion to overtures from unknown individuals. We will then experiment
with approaches to producing a bot capable of conducting dialog according to this theory. Of
greatest interest are approaches that attempt to backfit neural chit-chat models trained on large
dialog corpora to the domain theory. Experience with NLP research, ideally in NL dialogue is
required. Familiarity and experience with neural approaches to language modeling is a plus.
Twitter Event Processing. The aim of this project is to assess the impact of events in the news on
discussions found in social media, particularly twitter. The intern will develop algorithms to resolve
references to events in twitter to news articles and to classify and extract relevant details about
those events, as reported in the news, for the purpose of assessing their impact in the form of
twitter buzz. The intern will also create methods for assessing the magnitude of response to an
article, ideally with reference to inferred demographic or personality factors of the responding
users. Experience with NLP research, social media, machine learning is required. Familiarity with
relevant social science research is of interest.
Knowledge Base Population. The goal of this project is to develop a natural language processing
system that can automatically create a formal knowledge base from unstructured textual
information found on the Web. The intern will integrate existing SRI text analytic software
components, including text classification, language modeling, entity set expansion, term similarity
assessment, and information extraction into a coherent active learning architecture for knowledge
base population. Experience in NLP, machine learning, and Java programming is required.
Experience in linguistics and knowledge representation is a bonus.
Inline Evidence for Automated Writing Assessment. The goal of this project is to enhance an
existing SRI automated essay scoring software platform with a new capability for highlighting and
describing evidence of the system's decisions. The intern will investigate methods for determining
the influence of extracted text features on machine learning module predictions and pair this with
the development of a new capability for tracing the provenance of features back to the portions of
text from which they originate, and describing that provenance to end users. This will enhance
automated student feedback with more practical explanations that students can use to improve
their writing. Experience in NLP, machine learning, and Java programming is required. Experience in
linguistics and writing assessment is a bonus.
Novel Deep Learning Architectures. The goal of this project is to develop network construction and
training techniques with improved optimization behavior. We are currently conducting research on
the construction of adaptive networks that simultaneously model both empirical data and
formalized knowledge. The intern will study the impact of network properties and regularization
techniques on the convergence of optimization algorithms and quality of solutions. Experience with
network implementation, training and evaluation, either in Theano or TensorFlow, familiarity with
machine learning techniques, and experience working with large datasets is preferred.