With the growing deployment of pre-trained models like Transformers on cloud platforms, privacy concerns about model parameters and inference data are intensifying. Existing Privacy-Preserving Transformer Inference (PPTI) frameworks face the “impossible trinity” of balancing privacy, efficiency, and performance: Secure Multi-Party Computation (SMPC)-based approaches ensure strong privacy but suffer from high computational overhead and performance losses; Conversely, permutation-based methods achieve near-plaintext efficiency and accuracy but compromise privacy by exposing sensitive model parameters and intermediate results. Bridging this gap with a single approach presents substantial challenges, motivating the introduction of CENTAUR, a groundbreaking PPTI framework that seamlessly integrates random permutations and SMPC to address the “impossible trinity”. By designing efficient PPTI algorithms tailored to the structural properties of Transformer models, CENTAUR achieves an unprecedented balance among privacy, efficiency, and performance. Our experiments demonstrate CENTAUR’s ability to resist diverse data reconstruction attacks, achieve plaintext-level inference accuracy, and boost inference speed by 5.0~30.4 times, unlocking new possibilities for secure and efficient AI deployment.

Large language models (LLMs) have made significant breakthroughs in extracting useful information from conversation history to enhance the response in long-term conversations. Summarizing useful information from historical conversations has achieved remarkable performance, which, however, may introduce irrelevant or redundant information, making it difficult to flexibly choose and integrate key information from different sessions during memory retrieval. To address this issue, we propose a Fragment-then-Compose framework, a novel memory utilization approach for long-term open-domain conversation, called *FraCom*. To be specific, inspired by the concept of proposition representation from Cognitive Psychology, we first represent the conversation history as a series of predicates plus arguments for propositional representation to preserve key information useful for memory ("**Fragment**”). Then, we compose propositional graphs for the conversation history based on the connection between shared arguments ("**Compose**”). During retrieval, we retrieve relevant propositions from the graph based on arguments from the current query. This essentially allows for flexible and effective utilization of related information in long-term memory for better response generation towards a query. Experimental results on four long-term open-domain conversation datasets demonstrate the effectiveness of our *FraCom* in memory utilization and its ability to enhance response generation for LLMs.

Reinforcement Learning from Human Feedback (RLHF) is effective for aligning Large Language Models (LLMs) with human preferences. However, RLHF’s complex process limits its ability to continually learn human feedback, making it impractical for real-world applications where the deployed model continuously receives feedback from users. The non-RL-based method, such as Direct Preference Optimization (DPO), is not primitively favorable for Continual Learning (CL). We observe that when combined with Experiment Relay (ER) for CL, DPO tends to significantly widen the gap in the probability of human-preferred and dispreferred responses. Consequently, this diminishes the diversity in model generation, potentially leading to model collapse. To overcome the above challenges, we propose the Continual Optimal Policy Regularization (COPR), a novel non-RL offline method to convert the historical optimal policies into optimization constraints when continually learning new preferences. We first derive a moderate reward function from the pairwise ranking loss and then use the moderate reward to calculate a new sampling distribution to construct novel learning objectives and constraints. We also provide formal proof of the learnability of COPR. The experimental results show that COPR outperforms strong CL baselines on our proposed benchmark, in terms of reward-based, GPT-4 evaluations and human assessment.

In active retrieval (AR), large language models (LLMs) need first assess whether they possess knowledge to answer a given query, to decide whether to invoke a retrieval module. Existing methods primarily rely on training classification models or using the confidence of the model’s answer to determine knowledge boundaries. However, training-based methods may have limited generalization, and our analysis reveals that LLMs struggle to reliably assess whether they possess the required information based on their answers, often biased by prior cognitive tendencies (e.g., tokens’ semantic preferences). To address this, we propose Debiased Historical In-Context Learning (DH-ICL) to identify knowledge boundaries in AR. DH-ICL aims to reframe this self-awareness metacognitive task as a structured pattern-learning problem by retrieving similar historical queries as high-confidence in-context examples to guide LLMs to identify knowledge boundaries. Furthermore, we introduce a historical bias calibration strategy that leverages deviations in the model’s past response logits to mitigate cognitive biases in its current knowledge boundary assessment. Experiments on four QA benchmarks show that DH-ICL achieves performance comparable to full retrieval on LLaMA with only half the number of retrievals, without any additional training.

Prompt-based methods have shown their efficacy in transferring general knowledge within pre-trained language models (PLMs) for low-resource scenarios. Typically, prompt-based methods convert downstream tasks to cloze-style problems and map all labels to verbalizers.However, when applied to zero-shot entity and relation extraction, vanilla prompt-based methods may struggle with the limited coverage of verbalizers to labels and the slow inference speed. In this work, we propose a novel Discriminate Soft Prompts (DSP) approach to take advantage of the prompt-based methods to strengthen the transmission of general knowledge. Specifically, we develop a discriminative prompt method, which reformulates zero-shot tasks into token discrimination tasks without having to construct verbalizers.Furthermore, to improve the inference speed of the prompt-based methods, we design a soft prompt co-reference strategy, which leverages soft prompts to approximately refer to the vector representation of text tokens. The experimental results show that, our model outperforms baselines on two zero-shot entity recognition datasets with higher inference speed, and obtains a 7.5% average relation F1-score improvement over previous state-of-the-art models on Wiki-ZSL and FewRel.

With increasing concerns about data privacy, there is an increasing necessity of fine-tuning pre-trained language models (PLMs) for adapting to downstream tasks located in end-user devices or local clients without transmitting data to the central server. This urgent necessity therefore calls the research of investigating federated learning (FL) for PLMs. However, large PLMs bring the curse of prohibitive communication overhead and local model adaptation costs for the FL system. To this end, we investigate the parameter-efficient tuning (PETuning) of PLMs and develop a corresponding federated benchmark for four representative PETuning methods, dubbed FedPETuning. Specifically, FedPETuning provides the first holistic empirical study of representative PLMs tuning methods in FL, covering privacy attacks, performance comparisons, and resource-constrained analysis. Intensive experimental results have indicated that FedPETuning can efficiently defend against privacy attacks and maintains acceptable performance with reducing heavy resource consumption. The open-source code and data are available at https://github.com/SMILELab-FL/FedPETuning.