Home
About
People
Publications
Seminars
- Seminar Calendar
- Seminar Videos
Workshops
HLT Masters
Resources

The Economist features work by CLSP Ph.D. student Nathaniel R. Robinson

July 31, 2024

Categories: Uncategorized

Previous Article

Subscribe

Subscribe to the CLSP Seminars Mailing List

Upcoming Seminars

12:00 pm Findings of the First Shared Tas... @ Hackerman Hall 226

Findings of the First Shared Tas... @ Hackerman Hall 226

Nov 17 @ 12:00 pm – 1:15 pm

Abstract Efforts towards better machine translation (MT) for Creole languages have historically been isolated, due to Creole languages’ geographic and linguistic diversity. However, most speakers of Creole languages stand to benefit from improved MT for[...]

12:00 pm peerRTF: Robust MVDR Beamforming... @ Hackerman Hall B17

peerRTF: Robust MVDR Beamforming... @ Hackerman Hall B17

Nov 24 @ 12:00 pm – 1:15 pm

Abstract The Relative Transfer Function (RTF) is defined as the ratio of Acoustic Transfer Functions (ATFs) relating a source to a pair of microphones after propagation in the enclosure. Numerous studies have shown that beamformers[...]

Add

Add to Timely Calendar
Add to Google
Add to Outlook
Add to Apple Calendar
Add to other calendar
Export to XML

Seminar Videos

Paper Link: https://arxiv.org/abs/2505.22037

Abstract: Large language models (LLMs) are rapidly deployed in critical applications, raising urgent needs for robust safety benchmarking. We propose Jailbreak Distillation (JBDistill), a novel benchmark construction framework that "distills" jailbreak attacks into high-quality and easily-updatable safety benchmarks. JBDistill utilizes a small set of development models and existing jailbreak attack algorithms to create a candidate prompt pool, then employs prompt selection algorithms to identify an effective subset of prompts as safety benchmarks. JBDistill addresses challenges in existing safety evaluation: the use of consistent evaluation prompts across models ensures fair comparisons and reproducibility. It requires minimal human effort to rerun the JBDistill pipeline and produce updated benchmarks, alleviating concerns on saturation and contamination. Extensive experiments demonstrate our benchmarks generalize robustly to 13 diverse evaluation models held out from benchmark construction, including proprietary, specialized, and newer-generation LLMs, significantly outperforming existing safety benchmarks in effectiveness while maintaining high separability and diversity. Our framework thus provides an effective, sustainable, and adaptable solution for streamlining safety evaluation.

Jailbreak Distillation: Renewable Safety Benchmarking (EMNLP 2025)

Paper Link: https://arxiv.org/abs/2504.16046

Abstract: The exposure of large language models (LLMs) to copyrighted material during pre-training raises concerns about unintentional copyright infringement post deployment. This has driven the development of "copyright takedown" methods, post-training approaches aimed at preventing models from generating content substantially similar to copyrighted ones. While current mitigation approaches are somewhat effective for average-case risks, we demonstrate that they overlook worst-case copyright risks exhibits by the existence of long, verbatim quotes from copyrighted sources. We propose BloomScrub, a remarkably simple yet highly effective inference-time approach that provides certified copyright takedown. Our method repeatedly interleaves quote detection with rewriting techniques to transform potentially infringing segments. By leveraging efficient data sketches (Bloom filters), our approach enables scalable copyright screening even for large-scale real-world corpora. When quotes beyond a length threshold cannot be removed, the system can abstain from responding, offering certified risk reduction. Experimental results show that BloomScrub reduces infringement risk, preserves utility, and accommodates different levels of enforcement stringency with adaptive abstention. Our results suggest that lightweight, inference-time methods can be surprisingly effective for copyright prevention.

Certified Mitigation of Worst-Case LLM Copyright Infringement (EMNLP 2025)

Speaker: Zhouxiang Fang

Paper link: https://arxiv.org/abs/2504.19395

Abstract: Recent works have suggested that In-Context Learning (ICL) operates in dual modes, i.e. task retrieval (remember learned patterns from pre-training) and task learning (inference-time ''learning'' from demonstrations). However, disentangling these the two modes remains a challenging goal. We introduce ICL CIPHERS, a class of task reformulations based on substitution ciphers borrowed from classic cryptography. In this approach, a subset of tokens in the in-context inputs are substituted with other (irrelevant) tokens, rendering English sentences less comprehensible to human eye. However, by design, there is a latent, fixed pattern to this substitution, making it reversible. This bijective (reversible) cipher ensures that the task remains a well-defined task in some abstract sense, despite the transformations. It is a curious question if LLMs can solve tasks reformulated by ICL CIPHERS with a BIJECTIVE mapping, which requires ''deciphering'' the latent cipher. We show that LLMs are better at solving tasks reformulated by ICL CIPHERS with BIJECTIVE mappings than the NON-BIJECTIVE (irreversible) baseline, providing a novel approach to quantify ''learning'' in ICL. While this gap is small, it is consistent across the board on four datasets and six models. Finally, we examine LLMs' internal representations and identify evidence in their ability to decode the ciphered inputs.

ICL Ciphers: Quantifying "Learning’’ in In-Context Learning via Substitution Ciphers-EMNLP 2025 main

Revising Transfer-Based MT in a Phrase-Based SMT Framework - Stefan Riezler - 2005

🎬 JSALT 2025 – Closing Presentations Day 2 - afternoon
📍 Live from FIT, Brno University of Technology (Czech Republic)
📅 Friday, August 1st, 2025 - 13:30 CEST

!!! Join us on ZOOM, if you want to actively participate in the discussion !!!
Zoom link: https://cesnet.zoom.us/j/99972547651

📌 Join us for the final days of the JSALT 2025 Summer Workshop, hosted at the Faculty of Information Technology, Brno University of Technology!
This wrap-up event brings together 5 weeks of intensive research and collaboration in speech and language technology.

👇 Don’t be scared to ask questions in the chat and bring up discussion topics — we’ll be happy to get to them!

🕘 Schedule Overview (approx timestamps):
🔹13:30 – Team Presentation:
🎭 Advancing Expert-Level Reasoning and Understanding in Large Audio Language Models
Chairs: Ramani Duraiswami and Santosh Kesiraju
🔹14:40 – Stretch / Caffeine Break ☕
🔹14:50 – Continuation of Team Presentations
🔹16:00 – Open Discussion, Q&A, and Final Comments
🔹16:30 – Coda: Recognition of JSALT staff, and next steps (Sanjeev Khudanpur and Honza Cernocky)

#JSALT2025 #SpeechTech #ASR #TTS #LLM #ArtificialIntelligence #livefrombrno #FinalPresentationDay #WorkshopHighlights

[slides] AuGi: Advancing Expert-Level Reasoning and Understanding in LALMs | JSALT Closing Day 2

Load More... Subscribe

Johns Hopkins University, Whiting School of Engineering

Center for Language and Speech Processing
Hackerman 226
3400 North Charles Street, Baltimore, MD 21218-2680

(410) 516-4237
clspadmin@lists.johnshopkins.edu

Whiting School of Engineering
Johns Hopkins University

Privacy Statement
Legal

© 2025 Johns Hopkins University

Center for Language and Speech Processing