Data Tech 2026

Selective adsorption on functionalized surfaces underpins many technologies in separation, sensing, and filtration, yet rational design remains challenging because adsorption depends on many coupled factors, including surface chemistry, molecular structure, and environment. Here, we present an inverse molecular design framework that integrates generative AI models with physics‑based adsorption free‑energy calculations to accelerate discovery of surface chemistries optimized for target capture.

In this approach, generative AI models explore large spaces of candidate functional groups and molecular motifs conditioned on desired adsorption behavior, while molecular simulations provide quantitative feedback through computed potentials of mean force. Adsorption is modeled using umbrella sampling to explicitly capture the free‑energy landscape between bound and unbound states, enabling robust estimation of relative binding affinities and selectivities. These physically grounded signals are used to guide and refine the generative model, closing the loop between hypothesis generation and validation.

By combining data‑driven exploration with first‑principles adsorption modeling, this workflow shifts molecular discovery from a forward, trial‑and‑error paradigm to a goal‑directed inverse design process. The result is a scalable, interpretable methodology for proposing experimentally actionable surface chemistries with improved confidence, reduced iteration time, and direct alignment to performance objectives.

Most campaign measurement frameworks rely on redemption-based metrics, which fail to distinguish true incremental impact from temporal demand shifting. We introduce the Longitudinal Measurement Framework (LMF), a causal inference–driven approach to quantify the long-term effects of campaigns on guest behavior.

LMF estimates changes in Guest Lifetime Value and churn using a difference-in-differences design over pre- and post-campaign periods. The framework produces treatment effects that account for partial incrementality, separating short-term transactional lift from sustained behavioral change. By incorporating counterfactual baselines and longitudinal tracking, LMF mitigates biases inherent in redemption-based approaches, including selection effects and pull-forward behavior.

We present the methodological foundations of LMF along with key considerations for large-scale implementation in real-world settings, emphasizing the need to move beyond point-in-time metrics toward longitudinal, causally grounded evaluation of marketing effectiveness.

Serving large language models efficiently in production is a hard problem. Traditional inference engines waste up to 80% of GPU memory through KV cache fragmentation and over-allocation, leading to poor throughput and high costs. This talk dives into PagedAttention, a key innovation that borrows virtual memory paging from operating systems as a solution to this and vLLM, the open-source engine built on top of it.

This talk will cover the theory, walk through using vLLM in practice, and look at benchmark results showing up to 24× throughput improvements. We'll close with a look at how vLLM has been rapidly adopted across the industry and why PagedAttention has become a foundational primitive in LLM serving.

Can multiple AI agents, built across different frameworks, owned by different teams, and trained on different data sources, come together to solve an enterprise-scale problem? That's the central question SimRetail, developed at Target Corporation and accepted at AAMAS 2026 (the International Conference on Autonomous Agents and Multiagent Systems), is built to explore.

Retail assortment planning is a complex, high-stakes decision that has traditionally required human merchandisers to manually synthesize signals across market trends, sales performance, vendor constraints, and consumer behavior. SimRetail reimagines this as a multi-agent collaboration challenge, where specialist agents spanning trend research, merchandising analytics, and vendor intelligence must coordinate reasoning across multiple decision cycles to produce a coherent, consumer-validated assortment. A persona-based scoring agent, powered by NVIDIA's Nemotron-Personas-USA dataset of over 181,000 synthetic consumer profiles, then evaluates the output against eight buyer archetypes, stress-testing whether the agents' collective judgment actually translates to real consumer resonance.

Attendees will walk away with practical insights into multi-agent system design, agent-to-agent communication protocols such as A2A, MCP, challenges and evaluation of various agentic AI frameworks, observability solutions, synthetic persona evaluation, and building explainable agentic pipelines.

Generative models generate objects (picture, animation, sound, etc) by iteratively converting noise into data. This is done by the simulation of ordinary or stochastic differential equations (ODEs/SDEs). The two techniques allow us to construct, train, and simulate such ODEs/SDEs with deep neural networks.

In this talk I will introduce the two methods and will give simple examples.