Papers by Alessandra Russo
Red-Bandit: Test-Time Adaptation for LLM Red-Teaming via Bandit-Guided LoRA Experts (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to audit Large Language Models (LLMs) lack mechanisms to efficiently adapt to model-specific vulnerabilities at inference. |
| Approach: | They propose a red-teaming framework that adapts online to identify and exploit model failure modes under distinct attack styles. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on AdvBench and HarmBench, while generating more human-readable adversarial prompts (lower perplexity). |
Numerical reasoning in machine reading comprehension tasks: are we there yet? (2021.emnlp-main)
Copied to clipboard
| Challenge: | Numerical reasoning based machine reading comprehension models have achieved near-human performance on a variety of benchmarks, but are they capable of learning to reason? |
| Approach: | They propose to use a DROP benchmark to measure machine reading comprehension and investigate models that have achieved near-human performance over standard metrics. |
| Outcome: | The DROP benchmark has inspired the design of specialized BERT and embedding the results into a specialized model. |
Discrete Reasoning Templates for Natural Language Understanding (2021.eacl-srw)
Copied to clipboard
| Challenge: | Existing approaches to reasoning over multiple parts of a passage provide little evidence of their reasoning process, especially with regards to why specific operands are chosen for a reasoning task. |
| Approach: | They propose a method that decomposes complex questions into subquestions that can take advantage of single-span extraction models and derives the final answer according to instructions in a predefined reasoning template. |
| Outcome: | The proposed approach is interpretable and requires little supervision while competing with the state-of-the-art models. |
Towards preserving word order importance through Forced Invalidation (2023.eacl-main)
Copied to clipboard
| Challenge: | Recent studies show pre-trained language models are insensitive to word order . performance on NLU tasks remains unchanged even after permuting the word . |
| Approach: | They propose a simple approach called Forced Invalidation to force the model to identify permuted sequences as invalid samples. |
| Outcome: | The proposed approach significantly improves the sensitivity of the models to word order on English NLU and QA tasks over BERT-based and attention-based models over word embeddings. |
Learning and Enforcing Context-Sensitive Control for LLMs (2025.acl-srw)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have been able to achieve syntactic correctness but ensuring semantic validity requires additional mechanisms. |
| Approach: | They propose a framework that automatically learns context-sensitive constraints from LLM interactions through syntactic exploration and constraint exploitation. |
| Outcome: | The proposed framework outperforms larger models and state-of-the-art models in learning and generation of large LLMs. |