Tech geek, sunset seeker, and binge TV watcher. Tech marketing @Qualcomm and living in San Diego. @UCBerkeley and @UCLAAnderson alum. All opinions are mine.
AI is moving to the edge, and it's driving one of the biggest upgrade cycles the industry has seen.
At #COMPUTEX2026, I shared how @Qualcomm is uniquely positioned across the entire compute continuum, from sub-2 milliwatts to 2,000 watts. We also announced Dragonfly, our new family of data center products, demonstrating that we have a Dragon for every market.
More at Investor Day, June 24.
#QCOMInvestorDay
This week we introduced @Qualcomm Dragonfly, our new brand for data center products. More to come at our Investor Day on June 24.
Details: https://t.co/klw9eNSZsC
20+ demos. 17 accepted papers. Breakthrough ideas shaping what’s next.
At #NeurIPS2025, #AI innovation and Qualcomm moved in lockstep. Take a look at the highlights:
.@Qualcomm’s research shown at #NeurIPS2025 from generative AI to multimodal reasoning, and foundational ML—is shaping the future of edge AI.
https://t.co/uTLBOjHSNG
🚀 ICYMI, @Qualcomm#AI Research pushed the boundaries of AI at #NeurIPS2025!
From efficient model scaling to cutting-edge generative techniques, discover how we’re shaping the future of AI.
🔗 Read more: https://t.co/KvQEETEo1S
#AI#MachineLearning#QualcommAI
In this episode, we’re joined by Munawar Hayat, researcher at @QCOMResearch, to discuss a series of papers presented at NeurIPS 2025 focusing on multimodal and generative AI. We dive into the persistent challenge of object hallucination in Vision-Language Models (VLMs), why models often discard visual information in favor of pre-trained language priors, and how his team used attention-guided alignment to enforce better visual grounding. We also explore a novel approach to generalized contrastive learning designed to solve complex, composed retrieval tasks—such as searching via combined text and image queries—without increasing inference costs. Finally, we cover the difficulties generative models face when rendering multiple human subjects, and the new "MultiHuman Testbench" his team created to measure and mitigate issues like identity leakage and attribute blending. Throughout the discussion, we examine how these innovations align with the need for efficient, on-device AI deployment.
🗒️ For the full list of resources for this episode, visit the show notes page: https://t.co/5t89VH6oP1.
📖 CHAPTERS
===============================
00:00 - Introduction
04:35 - Physics-aware generation
06:56 - Challenges in physics-based visual generation
10:26 - Attention Guided Alignment in Vision Language Models paper
15:30 - Injecting visual tokens with cross-attention
18:45 - Computational performance during training
20:06 - Cross-attention for reducing squared complexity for longer context
21:05 - Benchmarks
23:16 - Hallucination and failure modes in VLMs
29:45 - Generalized Contrastive Learning (GCL) paper
38:01 - Retrieval on mobile devices
39:56 - ComGeneralized contrastive learning
40:46 - Benchmarks
41:54 - MultiHuman Testbench pape
49:33 - Efficiency against MultiHuman Testbench
51:33 - Qualcomm NeurIPS papers and demos
Interested in the latest multimodal generative #AI research? Check out this just-released @twimlai podcast to learn about @Qualcomm AI Research's latest contributions at @NeurIPSConf 2025.
https://t.co/Jy5Fa4BIoe
From edge to cloud, Qualcomm #AI Research is shaping what’s next in smarter technologies.
At #NeurIPS2025, we’re showcasing breakthroughs in generative AI, multimodal reasoning, and efficient neural networks.
Explore our work here: https://t.co/OZ9PFakshu
At #NeurIPS2025, we’ll demo our recent mobile model: 1.5B DiT generating 49×640×1024 videos in 8s.
After 2 years of research, we learned frontier video models can be distilled at little loss to run efficiently at the edge.
White paper: https://t.co/MMbkNfIc1h
Model: To come
At #NeurIPS2025, we’ll demo our recent mobile model: 1.5B DiT generating 49×640×1024 videos in 8s.
After 2 years of research, we learned frontier video models can be distilled at little loss to run efficiently at the edge.
White paper: https://t.co/MMbkNfIc1h
Model: To come
🎉We won the honorable runner up award at RealADSim challenge for generating novel driving scenes:
https://t.co/PYBVKddsCq
TLDR: Our solution complement Gaussian Splatting with a diffusion model to generate the missing details: https://t.co/LZnSzQvgCA
@Qualcomm#AI Research joins the @twimlai podcast to discuss high-efficiency diffusion models for on-device image generation and editing.
Learn how we're pushing the boundaries of generative AI to enable stunning visuals—right from your smartphone.
Listen now: https://t.co/Wf5TuBhzGN
.@humain is targeting 200 megawatts of Qualcomm AI200 and AI250 rack solutions, starting in 2026, to deliver high-performance #AI inference services in the Kingdom of Saudi Arabia and globally.
Learn more here: https://t.co/T9RQw5R7iK