Integrated Calendar

Transformer-based models have revolutionized natural language processing (NLP) and significantly improved various NLP tasks. However, many researchers make implicit assumptions about their training setups, assuming that the train and test sets are drawn from the same distribution. This assumption can limit the applicability of these models across different languages and domains.
The high cost of training state-of-the-art NLP models using various languages and domains has resulted in training them for only a subset of languages and domains, leading to a significant performance gap in excluded domains and languages. This performance gap marginalizes many individuals from accessing useful models.

This talk will address the challenges, approaches, and opportunities for democratizing NLP across different languages and domains.
Finally, we will explore future directions for making these models accessible to a broader audience.

We show that every language in NP has an Interactive Oracle Proof (IOP) with inverse polynomial soundness error and small query complexity. This achieves parameters that surpass all previously known PCPs and IOPs. Specifically, we construct an IOP with perfect completeness, soundness error 1/n, round complexity O(loglog n), proof length poly(n) over an alphabet of size O(n), and query complexity O(loglog n). This is a step forward in the quest to establish the sliding-scale conjecture for IOPs (which would additionally require query complexity O(1)).

Our main technical contribution is a emph{high-soundness small-query} proximity test for the Reed--Solomon code. We construct an IOP of proximity for Reed--Solomon codes, over a field F with evaluation domain L and degree d, with perfect completeness, soundness error (roughly) max{1-delta , O(
ho^{1/4})}$ for delta-far functions, round complexity O(loglog d), proof length O(|L|/
ho) over F, and query complexity O(loglog d)

Daniel Harari 1, Ron Rotkopf 2, Shlomit Reich-Zeliger 3, Mathumathi Krishnamohan 1, Alona Dov 1, Vladislav Volchinski 1 and Gideon Schreiber 1
1. Dept. Biomolecular Sciences, The Weizmann Institute of Science, Rehovot, Israel
2. Bioinformatics Unit, The Weizmann Institute of Science
3. Dept. Immunology, The Weizmann Institute of Science

Transcriptomic analysis of tumor biopsies from metastatic melanoma patients (SKCM-TCGA) demonstrates a profound survival advantage for approximately one-third of patients who exhibit the highest levels of intratumoral type I Interferon (IFN-I) signaling (Hazard Ratio 0.36; Pr: 4E-06), these which coincide with a sharp increase in transcripts indicating infiltration of CD4-T, CD8-T, B-cells and Macrophages to the tumors. Pathway analysis furthermore demonstrates that these patients exhibit T- and B-cell activation and a Th-1 response. To test if IFN-I signaling is central to and not simply correlating with these associated factors, we employed an adeno-associated virus (AAV) delivery system, locally expressing mouse IFNβ in B16F10 cells grafted into congenic C57BL/6 mice, this a known cold tumor model particularly hard to treat. Whereas anti-PDL1 monotherapy had no response in this model, combination therapy slowed, and in some cases cleared the mice of tumors. AAV-IFNβ monotherapy alone can slow but will not cure the mice. In sharp contrast to localized IFN delivery, systemic IFN therapy showed no beneficial effects in slowing tumor growth. To examine this more deeply, we injected the mice bilaterally with B16F10 tumors, where one tumor received AAV- IFNβ and the contralateral tumor received control. Both the injected and contralateral tumors nevertheless demonstrated a large reduction in tumor size, this effect lost when repeated using IFNAR2 knockout mice. Furthermore both the IFN-treated and contralateral tumors exhibited a large increase in CD3+CD4+ and CD3+CD4- lymphocytes. We submit that enforcing localized IFN-I signaling to a tumor in melanoma can drive immune cell infiltration, with the potential to elicit a systemic immune response, and possibly even cure, particularly when used in combination with immune checkpoint inhibitors.

In real-world supervised learning problems, accurate and trustworthy labels are often elusive, with label noise being a pervasive challenge. In this talk, we will delve into the inherent robustness of conformal prediction---a powerful tool for quantifying predictive uncertainty---to label noise. We will address both regression and classification problems and characterize how and when we can generate uncertainty sets that include the true labels that are hidden from us. By navigating between theory and practice, we will showcase the conservative coverage of clean ground truth labels achieved by employing conformal prediction with noisy labels and commonly used score functions, except in adversarial cases.

Foundation models that connect vision and language have recently shown great promise for a wide array of tasks such as text-to-image generation. Significant attention has been devoted towards utilizing the visual representations learned from these powerful vision and language models. In this talk, I will present an ongoing line of research that focuses on the other direction, aiming at understanding what knowledge language models acquire through exposure to images during pretraining. We first consider in-distribution text and demonstrate how multimodally trained text encoders, such as that of CLIP, outperform models trained in a unimodal vacuum, such as BERT, over tasks that require implicit visual reasoning. Expanding to out-of-distribution text, we address a phenomenon known as sound symbolism, which studies non-trivial correlations between particular sounds and meanings across languages and demographic groups, and demonstrate the presence of this phenomenon in vision and language models such as CLIP and Stable Diffusion. Our work provides new angles for understanding what is learned by these vision and language foundation models, offering principled guidelines for designing models for tasks involving visual reasoning.

Bio:
Hadar Averbuch-Elor is an Assistant Professor at the School of Electrical Engineering in Tel Aviv University. Before that, Hadar was a postdoctoral researcher at Cornell-Tech. She completed her PhD in Electrical Engineering at Tel-Aviv University. Hadar is a recipient of several awards including the Zuckerman Postdoctoral Scholar Fellowship, the Schmidt Postdoctoral Award for Women in Mathematical and Computing Sciences, and the Alon Fellowship for the Integration of Outstanding Faculty. She was also selected as a Rising Star in EECS in 2020. Hadar's research interests lie in the intersection of computer graphics and computer vision, particularly in combining pixels with more structured modalities, such as natural language and 3D geometry.