Schedule — LIGHT

08:30–08:45 Welcome & Workshop Overview

Opening remarks and overview of LIGHT

08:45–09:20 Opening Keynote

Jezabel R. García & Antonio Tiene — Multiverse Computing

09:20–10:20 Research Session I — Compression, Distillation & Efficient Model Design

#12 — Small Models Are Views, Not Artifacts: The Case for Random-Access Compressed Model StorageRitwick Ghosh #22 — Don't Skip the Baseline: A Controlled Comparison of Dynamic Expert Skipping in MoEsJakob Steimle, Thomas Elsken, Daniel Ehrhardt, Lukas Rinder, Daniel Cremers, Xi Wang #35 — XMerge: Cross-Axis Selection and Reconstructive Layer Merging for LLM Depth CompressionJundong Hu, Shekar Ramachandran #36 — Same Score, Different Model: Two Regimes of Quantization Damage Below and Above the Accuracy PlateauRaj Shah

10:20–10:40 Coffee Break

10:40–11:15 Panel — Can Small Models Outperform Large Models in Real-World Environments?

Panel participants to be announced

11:15–12:15 Research Session II — Trustworthy, Reliable & Verifiable Small Models

#9 — When BitsAndBytes Changes the Evaluator: Reliability of 8/4-Bit LLM JudgesMann Patira #51 — A Guarantee That Costs One Bit: Certified Few-Shot Prompt Adaptation for Deployable Vision–Language ModelsKeisuke Yokota #56 — My Graph Doesn’t Help! Analyzing Causal Graph Construction, Representation, and Utility in LLMsAman Syed, Benjamin Li #60 — Towards Trustworthy Compact Judges: What Human Labels Can and Cannot Fix in Compact AI JudgesSuchir Salhan, Mingchuan Zou, Paula Buttery

12:15–13:15 Lunch Break

13:15–13:50 Afternoon Keynote

Elisabetta Farella — FBK

13:50–14:50 Research Session III — Small Models in Real-World Systems & Applications

#28 — Small Talk: Literature Review with Locally-Hosted Small Language ModelsTimothy Joseph Welch, Isabella Illari, Bryan Edward Kaiser #33 — Explainability-Guided History Distillation for Efficient Sequential RecommendationAkash Pandey, Addrish Roy, Kanisha Shah, Akshat Shreemali, Dwipam Katariya, Swapnil Shinde, Amanda Ding, Kalanand Mishra, Pranab Mohanty #68 — Same Text, Different Prediction: Serving-Context Nondeterminism in Text ClassifiersSanthosh Kumar Kasa, Siva Rajesh Kasa, Sumit Negi #77 — A 127K-Parameter Cell in a Fixed Horner Program: Exactness, Compression, and a Verified Scaffold for Cross-Prime Modular MultiplicationRob Sneiderman

14:50–15:20 Coffee Break

15:20–15:50 Industry Session — Lightweight AI in Production

#30 — From Clicks to Intent: Distilling LLM Intent Understanding into Compact, Deployable Models in Financial ServicesDianjing Fan, Yao Li, Kyaw Hpone Myint, Dwipam Katariya, Alexandre G. R. Day, Pranab Mohanty, Giri Iyengar #65 — On-device High-Resolution Image Editing with Hallucination-Aware Loss and Adaptive TilingYoung D. Kwon, Abhinav Mehrotra, Malcolm Chadwick, Alberto Gil Couto Pimentel Ramos, Sourav Bhattacharya

15:50–16:45 Poster & Networking Session

All accepted papers; poster discussion, demos and networking

16:45–17:00 Closing Discussion & Takeaways — What Makes a Small Model Actually Better?

Final discussion and takeaways

Tentative workshop agenda