Join our in-person meetup to hear talks from experts on cutting-edge topics across AI, ML, and computer vision.

Generative AI has demonstrated ability to produce sounds and images, works of art. Can algorithms be trained to comprehend affinity for creatives? Part of the answer, at least, is yes. It turns out affinity for music is predictable from its sound. We engage the full Mystery Science Theater 3000: how do we begin teaching models to comprehend and critique full creative works?

Generative AI has made remarkable strides in producing photorealistic images, videos, and multimodal content. Yet, aligning these generations with human users, while ensuring spatial coherence, logical consistency, and deployment scalability, remains a major challenge, especially for real-world media platforms. In this talk, I will present our recent progress in enhancing the reasoning and control capabilities of image/video generative models, structured around four key pillars:

1. Efficiency & Scalability — with systems like ECLIPSE and FlowChef;
2. Control & Editing — including Lambda-ECLIPSE and RefEdit;
3. Reliability & Security — through efforts such as SPRIGHT, REVISION, R.A.C.E., and WOUAF;
4. Evaluation & Metrics — via benchmarks/metrics like VISOR, ConceptBed, TextInVision, and VOILA.

Together, these contributions outline a cohesive vision for building controllable, robust, and scalable generative systems, core to advancing personalization, understanding, and automated content workflows in media streaming and beyond.

Diffusion models have achieved impressive performance in generating high-quality and diverse synthetic data. However, their success typically assumes a class-balanced training distribution. In real-world settings, multi-class data often follow a long-tailed distribution, where standard diffusion models struggle -- producing low-diversity and lower-quality samples for tail classes. While this degradation is well-documented, its underlying cause remains poorly understood. We investigate the behavior of diffusion models trained on long-tailed datasets and identify a key issue: the latent representations (from the bottleneck layer of the U-Net) for tail class subspaces exhibit significant overlap with those of head classes, leading to feature borrowing and poor generation quality. Importantly, we show that this is not merely due to limited data per class, but that the relative class imbalance significantly contributes to this phenomenon. To address this, we propose COntrastive Regularization for Aligning Latents (CORAL), a contrastive latent alignment framework that leverages supervised contrastive losses to encourage well-separated latent class representations. We highlight the effect of CORAL for many large imbalanced datasets.

This talk shows how researchers can use FiftyOne to make their work more visible and impactful. We will cover practical ways to showcase datasets, models, and results, connect research tools, and share work with a broader community. The goal is to help researchers extend the reach of their research beyond papers and into real, reusable workflows.

This talk explores how agentic AI is less about solving predefined problems and more about designing systems—much like creating a game. Instead of following fixed procedures, AI models learn strategies within structured environments, adapting through feedback and iteration.

Drawing parallels to games like Tetris, the talk highlights how success comes not from a single correct solution, but from discovering effective strategies within given constraints. It also emphasizes that AI behavior doesn’t live solely in the model, but emerges from the entire system—including data, training processes, and defined objectives.

This talk also touches on the ethical implications of this perspective, showing how unexpected or undesirable AI behavior is often a result of how the “game” was designed, rather than a failure of the model itself.

Ultimately, this talk reframes AI development as a design challenge: shaping the systems and conditions that produce intelligent behavior.

This talk explores the fine line between creativity and hallucination in generative models, revealing how both stem from the same underlying processes. It breaks down why models sometimes drift from user intent and presents practical strategies to improve output accuracy.

From techniques like classifier-free guidance and contrastive decoding to reasoning before generation, the video highlights how alignment can be strengthened without advanced training. It also introduces an accessible, iterative workflow using vision-language models and LLMs to evaluate and refine outputs.

Ultimately, this approach empowers everyday users to better control generative AI—extending beyond images to code, visualizations, and interactive systems.

GamED.AI is a multi-agent AI platform that automates the generation of pedagogically grounded educational games. Current gamified learning tools like Kahoot and Quizlet rely on narrow quiz mechanics and lack curriculum-aligned content generation. GamED.AI addresses these gaps through a three-phase pipeline - Planning, Generation, and Assembly, which is orchestrated by specialized AI agents with human-in-the-loop (HITL) chat steering. The system leverages RAG-grounded, bloom’s taxonomy based pedagogy and deterministic validators to ensure content quality. In benchmark evaluations, GamED.AI achieves a 90% validation pass rate, produces games at ~$0.48 per game (13× cheaper than alternatives), and reduces token usage by 73% over 50 games generated. The platform supports diverse, multi-interaction game types that go beyond traditional quiz formats, delivering scalable, cost-effective, and reliable educational game generation. Additionally, GamED.AI offers seamless integration with existing Learning Management Systems, enabling educators to embed educational games directly into their institutional workflows with minimal setup.