קולוקוויום וסמינרים

Machine learning models play a key role for service providers looking to gain market share in consumer markets. However, traditional learning approaches do not take into account the existence of additional providers, who compete with each other for consumers. Our work aims to study learning in this market setting, as it affects providers, consumers, and the market itself.

We begin by analyzing such markets through the lens of the learning objective, and show that accuracy cannot be the only consideration. We then propose a method for classification under competition, so that a learner can maximize market share in the presence of competitors.

We show that our approach benefits the providers as well as the consumers, and find that the timing of market entry and model updates can  be crucial. We display the effectiveness of our approach across a range of domains, from simple distributions to noisy datasets, and show that the market as a whole remains stable by converging quickly to  an equilibrium.

In this seminar, I will discuss several fundamental challenges and limitations associated with high-perceptual-quality image restoration methods, and propose practical restoration and compression schemes. Specifically, I will first examine deterministic image restoration algorithms and show why striving for high output quality while maintaining consistency with the input measurements inevitably leads to algorithmic instability and vulnerability to adversarial attacks.

Secondly, since the perceptual quality and distortion of the reconstructions are typically at odds with each other, a key challenge in image restoration is to minimize the distortion under a constraint of perfect output quality. To address this optimization problem, I will introduce a novel algorithm that leverages a rectified flow model to approximate the optimal solution.

Finally, I will present an innovative generative approach based on pre-trained diffusion models, which produces high-quality image samples along with their losslessly compressed bit-stream representations. This new generative framework seamlessly extends to a variety of tasks, including image compression, compressed image restoration, compressed image editing, and more generally, any compressed conditional generation task.

As program synthesizers become integrated into IDEs, programmers combine synthesized code and manually written code within the same project. We therefore built a Programming-by-Example (PBE) synthesizer that documents the example specifications provided to it alongside the result snippet that satisfies them.

We also modified the IDE to treat these example scopes as localized tests for the code they surround, in case they or the code are edited. Unless strict limitations are imposed on how users can edit example scopes, and where they can call the synthesizer, scopes of example-specified code can wind up encompassing others.

The programmer can decide to send such a hierarchical scope to the synthesizer, to refactor the code, to automatically correct manually-written code, or simply to fix a mistake found in an example. State of the art PBE synthesizers cannot handle this hierarchical specification: synthesis will only consider the outer-most block, discarding the user intention contained in inner scopes. This can lead to undesired, overfitted programs.

To address this information loss we propose Spec Scooping, a syntax-guided technique to “scoop” out and preserve the intent from inner example scopes. We accompany Spec Scooping with a host of IDE features to help programmers edit example scopes and the code inside them, including an identification of when specifications contradict each other. Since Spec Scooping enriches the specifications sent to the synthesizer, it also requires modifications to the bottom-up Observational Equivalence synthesis algorithm.

We implement Spec Scooping in a tool ScooPy, including a development environment that supports scooping and an extended synthesizer.We evaluate ScooPy on 33 benchmarks based on SyGuS competition benchmarks.

Our results show that hierarchical specifications are generally more expressive, and that, compared to nonhierarchical specifications at the same level of expressiveness, scooping can provide a performance boost. We also performed two small-scale qualitative studies of ScooPy to gauge the benefits of attaching the specifications to a synthesis result and of users’ interaction with ScooPy’s development environment.