Understanding Multimodal User Generated Content in Social Media @ShareChat
Speaker: Vikram Gupta, Rishubh Parihar
Talk Details: https://offnote.substack.com/p/unders...
With millions of posts getting uploaded on ShareChat and Moj every day, holistically understanding this content is paramount for serving relevant content to our users and maintaining integrity of our platform by timely detection of offensive content like NSFW, violence, baits etc.
Since these posts are in the form of video, images, audio and text, it is important to design multimodal models which can interpret the relationship between the different modalities. Moreover, doing this at a massive scale calls for designing models which are fast, accurate, scalable and data-efficient. In this talk, we will discuss the techniques that we use at ShareChat and Moj to solve these problems at scale.
===
The OffNote Labs AI Talk Series brings you industry experts, researchers and practitioners, passionate to share their learnings and experience -- on innovating and building cutting-edge AI technology / systems which touch and influence the lives of billions of people on this planet.
Our talks are informal, a blend of traditional presentation / podcast, and the audience very technically engaged.
We take delight in unraveling the experience of research and innovation, and celebrate innovators who 'follow the problem', and make complex technology work in the wild.
==
Follow OffNote Labs on LinkedIn. / offnote
Watch previous talks and Subscribe on Youtube: https://bit.ly/31VTMHH
Sign up to our newsletter: http://offnote.substack.com
Read our research articles: / offnote
Web: https://offnote.co
==
00:00 Introduction
08:41 Agenda: Bharat Vs India
13:33 Sharechat - Solution for the language - first audience
14:26 Moj -India’s #1 Short Video App
15:33 Shift from Connection Oriented to AI first feeds
17:46 Core Verticals
21:33 Content Moderation
23:06 Content Moderation Policies
30:26 off-the-shelf Moderation Solutions
32:13 Deduplication
38:40 Cold Start
41:33 Solving Cold Start
55:06 Multimodal Content Understanding
55:46 Temporal Modelling
56:13 Feature Extraction
56:40 Modality Fusion
58:53 Multitask Training
59:06 Fast & Scalable Model
1:06:26 Key Frames Extraction
1:07:20 Fast & Scalable Models - Efficient Backbones
1:09:06 Temporal Shift Modules (TSM)
1:10:26 Fast & Scalable Models - Network Pruning and Weight quantization
1:13:46 Knowledge Distillation
1:21:33 Data-Efficient Models
1:30:00 Semi-supervised Learning
1:34:13 Transductive Learning
1:35:20 Inductive Learning
1:35:46 Semi-supervised Learning is Challenging
1:37:06 Temporal Ensembling
1:38:26 Mean Teacher
#multimodalai #contentmoderation #sharechat #moj #semisupervised #knowledgedistillation #temporalensembling #coldstart #bharatvsindia