We look forward to meeting you there.

Welcome to Black in AI 2020 Workshop

"The success of diagnostic and interventional medical procedures is deeply rooted in the ability of modern imaging systems to deliver clear and interpretable information. After raw sensor data is received by ultrasound and photoacoustic imaging systems in particular, the beamforming process is often the first line of software defense against poor quality images. Yet, with today’s state-of-the-art beamformers, ultrasound and photoacoustic images remain challenged by channel noise, reflection artifacts, and acoustic clutter, which combine to complicate segmentation tasks and confuse overall image interpretation. These challenges exist because traditional beamforming and image formations steps are based on flawed assumptions in the presence of significant inter- and intrapatient variations. In this talk, I will introduce the PULSE Lab’s novel alternative to beamforming, which improves ultrasound and photoacoustic image quality by learning from the physics of sound wave propagation. We replace traditional beamforming steps with deep neural networks that only display segmented details and structures of interest. Our pioneering image formation methods hold promise for robotic tracking tasks, visualization and visual servoing of surgical tool tips, and assessment of relative distances between the surgical tool and nearby critical structures (e.g., major blood vessels and nerves that if injured will cause severe complications, paralysis, or patient death)."

We present in this work a Deep Reinforcement Learning Based Sequence-to-1Sequence model for Natural Question Generation. The question Generation task2aims to generate questions according to a text that serves as context and the answer.3Generate a question is a difficult task because it requires first a well understanding4of the context and the relation with the provided answer, then requires the ability5to generate natural questions as humans. The question should be syntactically,6semantically correct, and correlated with the context and answer. Based on these7constraints we used first Attention models based on Transformers specifically8Google T5 to address the Task of Natural Language Understanding and Natural9Language Generation, then used an Evaluator formed by the mixture of the Cross-10Entropy loss function and a Reinforcement Learning Loss. The aim of this hybrid11evaluator is to drive the training by ensuring that the generated questions are12syntactically and semantically correct. To train our model we used the benchmark13dataset for Reading Comprehension of Text SQUAD. As an evaluation metric,14we use the new State-of-the-Art evaluation metric NUBIA that provides a great15indicator to measure the linguistic similarity between two sentences. There are16many use cases of Question Generation, especially in education it can be used to17improve Question Answering dataset, it can also be used to improve learning by18supporting the student self-taught and help the teachers to design exams.

Introductory hands-on courses such as our smartphone-based coding courses, SuaCode require a lot of support for students to accomplish learning goals. Offering assistance becomes even more challenging in an online course environment which has become important recently because of COVID-19. Hence, offering quick and accurate answers could improve the learning experience of students. However, it’s difficult to scale this support with humans when the class size is huge. A few works have developed virtual teaching assistants (TA). All of these TAs have focused on logistics questions, and none have been developed and evaluated using coding courses in particular. Also, they have used one language (e.g. English). Given the multilingual context of our students — learners across 38 African countries — in this work, we developed an AI TA — Kwame — that provides answers to students’ coding questions from our SuaCode courses in English and French. Kwame is a Sentence-BERT(SBERT)-based question-answering (QA) system that we trained and evaluated using question-answer pairs created from our course’s quizzes and students’ questions in past cohorts. It finds the paragraph most semantically similar to the question via cosine similarity. We compared the system with TF-IDF and Universal Sentence Encoder. Our results showed that SBERT performed the worst for the duration of 6 secs per question but the best for accuracy and fine-tuning on our course data improved the result.

Come meet our sponsors in gather town!

Fireside Chat with Dr. Ramon Amaro discussing their new book on Machine Learning, Sociogeny, and Race.

"Apple’s AI chief John Giannandrea recently said in an interview “I really honestly think there's not a corner of iOS or Apple experiences that will not be transformed by machine learning over the coming few years." The widespread deployment of AI across every layer of digital platforms may be one of the most important paradigm shifts for UI/UX designers since the introduction of the iPhone in 2007. User interfaces continue to be deeply reshaped by the various breakthroughs made in machine learning. We’re entering an era of AI-assisted UIs which involves new mechanisms, new principles, and new aesthetics. This talk will walk through a few prototypes and projects I’ve worked on (such as Google Art Selfie and ClipDrop) and explore how some interactions can be re-imagined with AI."

Text-to-image generation via Generative Adversarial Networks (GANs) is largely explored within image generation from captions. However, semantic exploration or integrating knowledge bases into image generation is uncharted, which is why we seek to generate images from ideas or concepts that are obscure to imagine. Thus, to understand the visualization of concepts, we synthesize GAN visuals from a semantic knowledge graph with meanings and understanding of words.

We present NUBIA, a methodology to build automatic evaluation metrics for text generation using only machine learning models as core components. A typical NUBIA model is composed of three modules: a neural feature extractor, an aggregator and a calibrator. We demonstrate an implementation of NUBIA which outperforms metrics currently used to evaluate machine translation, summaries and slightly exceeds/matches state of the art metrics on correlation with human judgement on the WMT segment-level Direct Assessment task, sentence-level ranking and image captioning evaluation. The model implemented is modular, explainable and set to continuously improve over time.

As in-person health appointments continue to be limited, the role of online health communities as outlets for the exchange of information and emotional support can only increase in prominence. Further analysis and comparisons of these communities can provide valuable insights to the medical community and the general public. In a case study of two online birth control communities, we use topic modeling and a user-trained classifier to show that contraceptive information is sought and shared differently in different online settings.

Come meet our sponsors in gather town!

We will hold mentorship roundtables with leaders in academia and industry to discuss a range of research topics and provide guidance on going to graduate school and academia as well as applying to jobs in industry.

Join us for a great poster session, jointly with other affinity groups.

Dr. Guimarães is an Adjunct IV Professor at Aeronautics Institute of Technology – ITA in Portuguese. Dr. Guimarães teaches experimental physics for future electrical, computational, structures for airports, mechanics for airplanes, aeronautics e aero spatial engineers, in their 1st and 2nd years. She used to research about semiconductors to make solar cells and infrared sensors, now she is researching about organic light emitting diodes – OLEDs. The idea is to use OLEDs to make teaching physics more interesting to high school students. She completed her PhD in Semiconductor Devices, at The University of Manchester Institute of Science and Technology, England. She also holds a MS in Applied Physics at São Paulo University – USP in Portuguese, Brazil. She is the counsel founder of afrobras - a non-governmental initiative, which created Zumbi dos Palmares University, the first Historically Black Brazilian University. She is also the City of São José dos Campos counsel to promote the Racial Equality, in São Paulo State. She is fellow of the Physics Racial Equity Work Group of the Brazilian Physics Association – SBF in Portuguese, and of the of the Black Researchers - ABPN in Portuguese. She travels around several states of Brazil talking about persuading girls to choose STEM as carrier, digital revolution in order to help them to choose their future. Advocating again racism and gender discrimination. Speeches to motivate victims of these crimes to fight for justice. Dr. Guimarães was born and raised in São Paulo, southeast of Brazil.

Authorship attribution on social media is difficult, especially on Twitter, where the posts have character limitations. The increased prevalence of disinformation on social media motivates us to study how much of an 'authorship footprint' is left on social media by celebrities vs. non-celebrities and various professional groups. In this talk, Dr. Siobahn Day Grady will discuss authorship within online social networks and present examples of different groups, their footprints on these networks, and machine learning algorithms performance in determining professional groups' authorship.