Projects and Writeups
(more to come...)
The concept of "Revealable Functional Commitments" involves a cryptographic method that allows a prover to commit to a secret function and later reveal certain aspects of its functionality without exposing the entire function. This is achieved through a "proof of reveal," which validates the consistency of a set of constraints with the committed secret function. Combining an algebraic holomorphic proof (AHP), a proof of function relation (PFR), and a proof of reveal, a secure revealable functional commitment scheme is constructed. The paper also introduces interactive protocols for properties of committed polynomials, which may be independently valuable
n-Party Private (NPP) circuits are those where disjoint subcircuits are held by multiple independent entities. In this project, efficient functional commitments are redesigned, along with their associated proofs of function relation, to work for this special class of functions. The result is a plug-and-play mechanism for parties to jointly compute functions where certain subroutines are proprietary, and the composite output is revealed in zero-knowledge.
​
(previously titled PACCMulE)
Class Conditional Adaptive Predictive Sets (CCAPS1) is an algorithm that provides finite-sample coverage guarantees for prediction sets from any black-box classifier, conditioned on inputs from any class. Unlike prior methods, CCAPS1 ensures coverage for each class, not just on average across the dataset. It is universally applicable, model-agnostic, and allows for class-specific coverage customization. Its effectiveness is empirically validated on standard datasets and various classifiers.
"The Cutting Edge" provides an in-depth exploration of fair division in cake-cutting, introducing a formal model to ensure envy-free distribution of a divisible resource. The study delves into the complexities of allocating both connected and disconnected pieces, presenting algorithms for different numbers of participants, and discussing bounds on query complexity. It addresses practical issues such as the desire for connected allocations, the challenges of bad cakes (negative valuations), and the intricacies of various cake shapes. The research highlights the current gaps and the potential for future discoveries in this fascinating area of mathematics and computer science.
