| Time | September 7th, 2026 @ RedHat Raanana |
|---|---|
| 9:00 | Registration, coffee & refreshments |
| 9:45 | Opening Session |
| 10:00 | Keynote: Tamar Eilam (IBM) |
| 11:00 | Break |
| 11:15 | Efficient and Robust Deep Learning – Session A |
| 12:30 | Break |
| 12:45 | AI Factory Systems – Session B |
| 13:35 | Lunch |
| 14:45 | Optimizing AI Systems at Scale – Session C |
| 16:25–17:30 | Poster Session and Reception |
The emergent ability of LLMs to reason, hypothesize, experiment, and generate code is forcing us to rethink not just how we develop code, but what we design, and how we continuously evolve entire software systems.
Modern systems are under increasing pressure to meet new requirements driven by a rapidly changing technological landscape such as new model architectures, hardware innovations, and shifting workloads. Yet their improvement remains fundamentally human-mediated, one decision at a time, and therefore largely reactive. At IBM Research we challenged ourselves with a question: can AI become the primary agent of system evolution, so that a deployed system observes its own behavior, hypothesizes improvements, experimentally validates them, and deploys, continuously and at machine speed?
In this talk I will describe the principles we uncovered, the tools we built, and the proof points we obtained across several systems domains: inference serving optimization, accelerator compute kernel generation, and a domain-specific storage system. I will situate this within the fast-evolving academic landscape and conclude with some key open questions.
Tamar Eilam received her Ph.D. in Computer Science from the Technion in 2000 and joined the IBM T.J. Watson Research Center as a Research Staff Member that same year. She was named an IBM Fellow in 2014 and her main research interest is the use of AI to manage the entire lifecycle of software systems including algorithmic discovery and optimization.
Talks
Efficient and Robust Deep Learning – Session A
Chair: Gala Yadgar (Technion)
- Highlight: Attention Needs Less Memory: From LLM Training to Long-Context Inference
Presented by: Malik Khalaf (Technion)
CoRR2025
Slides - Highlight: GASLITEing the Retrieval: Exploring Vulnerabilities in Dense Embedding-based Search
Presented by: Matan Ben Tov (Tel-Aviv University)
CCS2025 - Highlight: Efficient Optimization of Massively Overparameterized Deep Neural Networks
Presented by: Tom Tirer (Bar-Ilan University)
TMLR2026, NeurIPS 2025
Slides
AI Factory Systems – Session B
Chair: Eran Gilad (Regatta Data)
- What Will We Build with 100K AI Accelerators and how can You Participate
Guy Altagar (The National AI Program)
Slides - Beyond the Vector Retrieval Trap: Implementing True Episodic Agentic Memory
Guy Korland (FalkorDB)
Slides
Optimizing AI Systems at Scale – Session C
Chair: Anastasia Braginsky (RedHat)
- From Models to Manifests: Automating LLM Deployment Planning for llm-d on Kubernetes
Amit Oren (Red Hat)
Slides - The Evolution of AI: A Compute Story
Adi Fuchs (Nvidia)
Slides - AI Factory Fabric landscape and directions
Shahar Belkar (Toga Networks)
Slides - Highlight: LLM KV-cache: I/O-aware KV-cache Management
Erez Zadok (Stony Brook University)
HotStorage2026
Slides
Poster Session
Chair: Sarel Cohen (Reichman University)
For accepted posters, see
Venue
Infinity Tower, 15th floor (Reception) – 8 Hapnina St. Raanana
RedHat Raanana is a convenient 5 min walk from the train station.
For guest parking – register (after registering to ISAI) and then use the Plaza Tower guest parking entrance & lot (entrance from the HaBedolach St.)