Events / BOOST: BOlogna Orthogonal Summer Term

Ittay Eyal, Technion

Minicourse 1: Blockchains, Cryptocurrencies, and Decentralization

Decentralized blockchain systems are virtual machines available to anyone with Internet access. Moreover, they are permissioness distributed protocols, where any node can join to become one of the system operators. Their design utilizes tools from distributed computing, game theory, and cryptography. This combination gives rise to fundamental questions on how to achieve security and fairness without sacrificing performance; in turn, it raises open questions in each of the aforementioned fields. This short course will survey the theory and design of the different layers of blockchain systems and discuss some of the open questions.

Seminar 2: Time Is Money: Incentivized Causal Transaction Ordering

Front-running is a subtle and persistent problem for blockchains. A blockchain is a stateful virtual machine executing instructions called transactions. Users earn rewards by publishing functional transactions essential to the system. Attackers observe these transactions and publish their own ahead of the users’, seizing the reward and eroding users’ incentive to publish functional transactions. Preventing front-running means enforcing causality: If an attacker receives transaction txA and then publishes transaction txB, then txA must be ordered before txB. However, this causality is only observed by the attacker. Practical systems order transactions by bid amount, so transactions willing to pay more get executed first, but this only results in a bidding war eroding users’ rewards. Though numerous ordering approaches have been proposed, none achieves causality, leaving users vulnerable to front-running.

This talk will present PRECEDE, a mechanism-design approach that enforces transaction causality by removing the economic incentive to front-run. PRECEDE orders transactions by a power-weighted randomized lottery, whose winning probability grows super-linearly in the bid. The user’s strategy of publishing a transaction with a deterring bid forms an equilibrium where the attacker refrains from competing. Moreover, PRECEDE prevents the prominent sandwich attack, which relies on front-running. PRECEDE can be directly deployed in existing blockchains with a simple change to their transaction ordering mechanism.

Adrian Sampson, Cornell University

Minicourse 1: Modern Language Implementation

Compilers are the way that the computing ecosystem adapts to exploding diversity in both hardware and programming models. This short course is a hands-on introduction to the essential parts of a compiler “middle end,” which is where all the interesting work happens to represent, analyze, transform, and optimize programs. We will study program representations, local and global analyses, the data flow framework, and static single assignment form. The structure of the course will alternate between understanding algorithms and then implementing them to observe their effects on real code.

Seminar 2: Compiling Math to Hardware

Hardware accelerators are the main route to scaling computational efficiency past the end of Moore’s law. Specialized mathematical functions form the heart of many application-specific accelerators. This category of hardware gains its efficiency from customizing numeric representations and elementary function implementations beyond what is possible in a general-purpose CPU. However, designing these mathematical circuits from scratch is a multifaceted and complex engineering problem. A math-to-hardware compiler is a hypothetical toolchain that translates an abstract mathematical expression over real numbers into an efficient hardware circuit. This talk will discuss some recent steps toward realizing math-to-hardware compilers, including optimizing “libm”-like functions and safely composing different specialized number formats.

Antonio Orvieto, ELLIS Institute Tübingen, Max Planck Institute for Intelligent Systems

Seminar: Modern Optimization Theory Predicts LLM Training and Scaling Laws

Understanding how to adapt training hyperparameters (e.g., batch size, learning rate) as model size and data scale is crucial for developing future-generation AI systems. We study hyperparameter scaling laws for modern first-order optimizers through the lens of recent convergence bounds for methods based on the Linear Minimization Oracle (LMO), a framework that includes normalized SGD, signSGD (an Adam approximation), and Muon. Treating bounds in recent literature as proxies and minimizing them across different tuning regimes yield closed-form power-law schedules for learning rate, momentum, and batch size as functions of the iteration or token budget. Our analysis, holding model size fixed, recovers most insights and observations from the literature under a unified and principled perspective, with clear directions open for future research. Our results draw particular attention to the interaction between momentum and batch-size scaling, suggesting that optimal performance may be achieved with several scaling strategies.

Reference: https://arxiv.org/pdf/2603.15958