Large Language Models (LLMs) are susceptible to security and safety threats, such as prompt injection, prompt extraction, and harmful requests. One major cause of these vulnerabilities is the lack of an instruction hierarchy. Modern LLM architectures treat all inputs equally, failing to distinguish between and prioritize various types of instructions, such as system messages, user prompts, and data. As a result, lower-priority user prompts may override more critical system instructions, including safety protocols. Existing approaches to achieving instruction hierarchy, such as delimiters and instruction-based training, do not address this issue at the architectural level. We introduce the $\textbf{I}$nstructional $\textbf{S}$egment $\textbf{E}$mbedding (ISE) technique, inspired by BERT, to modern large language models, which embeds instruction priority information directly into the model. This approach enables models to explicitly differentiate and prioritize various instruction types, significantly improving safety against malicious prompts that attempt to override priority rules. Our experiments on the Structured Query and Instruction Hierarchy benchmarks demonstrate an average robust accuracy increase of up to 15.75\% and 18.68\%, respectively. Furthermore, we observe an improvement in the instruction-following capability of up to 4.1\% on AlpacaEval. Overall, our approach offers a promising direction for enhancing the safety and effectiveness of LLM architectures.

Wenxuan Zhou, Ravi Agrawal, Shujian Zhang, Sathish Reddy Indurthi, Sanqiang Zhao, Kaiqiang Song, Silei Xu, Chenguang Zhu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

The cascaded approach continues to be the most popular choice for speech translation (ST).This approach consists of an automatic speech recognition (ASR) model and a machine translation (MT) model that are used in a pipeline to translate speech in one language to text in another language.MT models are often trained on well-formed text and therefore lack robustness while translating noisy ASR outputs in the cascaded approach, degrading the overall translation quality significantly.We address this robustness problem in downstream MT models by forcing the MT encoder to bring the representations of a noisy input closer to its clean version in the semantic space.This is achieved by introducing a contrastive learning method that leverages adversarial examples in the form of ASR outputs paired with their corresponding human transcripts to optimize the network parameters.In addition, a curriculum learning strategy is then used to stabilize the training by alternating the standard MT log-likelihood loss and the contrastive losses.Our approach achieves significant gains of up to 3 BLEU scores in English-German and English-French speech translation without hurting the translation quality on clean text.

In this paper, we propose a new primal heuristic for the Linear Ordering Problem (LOP) that generates an integer feasible solution from the solution to the LP relaxation at each node of the branch-and-bound search tree. The heuristic first finds a partition of the set of vertices S into an ordered pair of subsets {S1,S2} such that the difference between the weights of all arcs from S1 to S2 and the weights of all arcs from S2 to S1 is maximized. It then assumes that all vertices in S1 precede all vertices in S2 thus decomposing the original problem instance into subproblems of smaller size i.e. on subsets S1 and S2. It recursively does so until the subproblems can be solved quickly using an MIP solver. The solution to the original problem instance is then constructed by concatenating the solutions to the subproblems. The heuristic is used to propose integer feasible solutions for the branch-and-bound algorithm. We also devise an alternate node selection strategy based on the heuristic solutions where we select the node with the best heuristic solution. We report the results of our experiments with the heuristic and the node selection strategy based on the heuristic.

Analysing datasets requires sophisticated techniques which can help to unearth interesting patterns. One approach is to mine multidimensional association rules from data. The traditional association rule mining relies on uniform support and confidence values which does not always yield interesting rules due to varied nature of data. This paper presents a novel approach to mine multidimensional association rules from dataset with varying support. The improved algorithm is being proposed to overcome missing aspect of tradition rule mining algorithms like Apriori.