• 18:00

    Welcome & Introduction

  • 18:05
    Avriel Epps-Darling

    Presentation: Do #BlackLivesMatter to YouTube?: Exploring YouTube Recommendation Pathways for Black Lives Matter Content After the Death of George Floyd

    Avriel Epps-Darling - Ph.D. Candidate - Harvard University

    Down arrow blue

    Do #BlackLivesMatter to YouTube?: Exploring YouTube Recommendation Pathways for Black Lives Matter Content After the Death of George Floyd

    Social media giants like Facebook, Twitter and YouTube have come out in support of Black Lives Matter and its mission since the death of George Floyd earlier this summer. But are their platforms supporting the movement for racial justice? In this talk, I will present preliminary findings of a content analysis of 845 videos collected from YouTube search queries of 55 keywords related to the BLM movement. The content that emerged from these queries reveals key racial and ideological disparities in YouTube’s search recommendations that may misalign with their recently published commitment to supporting the Black Lives Matter movement. Subsequent analyses on 3 million “Up Next” recommendations resultant from these search results (i.e., their recommendation pipelines) provide additional insights into the ways in which racial justice-related content is promoted on the platform.

    Avriel Epps-Darling is a Ph.D. candidate, Ford fellow, and Presidential Scholar at Harvard University. As a scholar, she has garnered numerous awards and honors including an invitation from the U.S. Department of Education to present her work for Congress in Washington D.C. and recognition as part of the top 10% of undergraduate social scientists in the world. Her previous research on algorithmic bias and music streaming compliments her foray into music making, where she took on the stage name King avriel. Her most recent musical project 'thesis' was released to critical acclaim, hailed as "prodigious" by the Huffington Post and featured in The New York Times, Vogue, Vice, and more. Today, her research, in partnership with organizations such as Spotify and Snap Inc, focuses on the intersection of algorithmic bias in content recommendation systems, and racial identity development.

    Twitter Linkedin
  • 18:20
    Wael AbdAlmageed

    Presentation: Multimedia Forensics In The Age GANs – The War Has Just Begun

    Wael AbdAlmageed - Research Associate Professor - USC Viterbi

    Down arrow blue

    Multimedia Forensics In The Age GANs – The War Has Just Begun

    Image and video manipulation have lately become an epidemic, mainly due to the ease-of-use and availability of multimedia editing tools, such as Photoshop and GIMP. Further, recent advances in deep generative modeling, such as Generative Adversarial Networks (GAN), gave birth to a new era of photo-realistic synthetic images. The widespread of manipulated and generated multimedia, as tools of misinformation, on social networking platforms has major societal consequences, including child safety, public opinion and democracy and criminal justice. The prevalence of fabricated multimedia necessitated a new wave of forensic computer vision methods for detecting and localizing image and video manipulations. In this talk I will give a brief overview of current challenges in multimedia forensics then I will then present two novel multimedia forensics methods developed at VISTA. First, I will present ManTraNet (Manipulation Tracing Network) a unified deep convolutional neural network architecture for detecting and localizing various types of image manipulations, such as copy-move and splicing. ManTraNet is considered state-of-the-art image manipulation detection framework and have been evaluated on multiple standard datasets. Second, I will present a state-of-the-art deep learning pipeline for detecting deepfake videos. Our pipeline demonstrated state-of-the-art evaluation results on FaceForensics++, the only large scale deepfake dataset. Finally, I will wrap up with conclusions and thoughts for future directions in multimedia forensics research.

    Dr. AbdAlmageed is a Research Associate Professor at Department of Electrical and Computer Engineering, and a research Team Leader and Supervising Computer Scientist with Information Sciences Institute, both are units of USC Viterbi School of Engineering. His research interests include representation learning, debiasing and fair representations, multimedia forensics and visual misinformation (including deepfake and image manipulation detection) and biometrics. Prior to joining ISI, Dr. AbdAlmageed was a research scientist with the University of Maryland at College Park, where he lead several research efforts for various NSF, DARPA and IARPA programs. He obtained his Ph.D. with Distinction from the University of New Mexico in 2003 where he was also awarded the Outstanding Graduate Student award. He has two patents and over 70 publications in top computer vision and high performance computing conferences and journals. Dr. AbdAlmageed is the recipient of 2019 USC Information Sciences Institute Achievement Award.

    Twitter Linkedin
  • 18:35
    Nazneen Fatema Rajani

    Presentation: Tailoring Word Embeddings for Gender Bias Mitigation

    Nazneen Fatema Rajani - Senior Research Scientist - Salesforce Research

    Down arrow blue

    Tailoring Word Embeddings for Gender Bias Mitigation

    Word embeddings derived from human-generated corpora inherit strong gender bias which can be further amplified by downstream models. Some commonly adopted debiasing approaches apply post-processing procedures that project pre-trained word embeddings into a subspace orthogonal to an inferred gender subspace. We discover that semantic-agnostic corpus regularities such as word frequency captured by the word embeddings negatively impact the performance of these algorithms. We propose a simple but effective technique that purifies the word embeddings against such corpus regularities prior to inferring and removing the gender subspace.

    Nazneen is a senior research scientist at Salesforce working on commonsense reasoning, interpretability, and robustness. She got her PhD in Computer Science from UT Austin in 2018. Several of her work (10+) has been published in top tier conferences like ACL, EMNLP, NACCL, and IJCAI including the work on debiasing word embeddings. Nazneen was one of the finalists for the VentureBeat Transform 2020 women in AI Research. She has given invited talks at various universities and conferences including Yale, UVA, TMLS, and Dreamforce.

    Twitter Linkedin
  • 18:50

    Audience Q&A

  • 19:00

    END OF WEBINAR

This website uses cookies to ensure you get the best experience. Learn more