As Large Language Models (LLMs) are increasingly deployed in specialized domains with continuously evolving knowledge, the need for timely and precise knowledge injection has become essential. Fine-tuning with paraphrased data is a common approach to enhance knowledge injection, yet it faces two significant challenges: high computational costs due to repetitive external model usage and limited sample diversity. To this end, we introduce LaPael, a latent-level paraphrasing method that applies input-dependent noise to early LLM layers.This approach enables diverse and semantically consistent augmentations directly within the model. Furthermore, it eliminates the recurring costs of paraphrase generation for each knowledge update. Our extensive experiments on question-answering benchmarks demonstrate that LaPael improves knowledge injection over standard fine-tuning and existing noise-based approaches. Additionally, combining LaPael with data-level paraphrasing further enhances performance.

Existing methods in domain generalization for Multimodal Sentiment Analysis (MSA) often overlook inter-modal synergies during invariant features extraction, which prevents the accurate capture of the rich semantic information within multimodal data. Additionally, while knowledge injection techniques have been explored in MSA, they often suffer from fragmented cross-modal knowledge, overlooking specific representations that exist beyond the confines of unimodal. To address these limitations, we propose a novel MSA framework designed for domain generalization. Firstly, the framework incorporates a Mixture of Invariant Experts model to extract domain-invariant features, thereby enhancing the model's capacity to learn synergistic relationships between modalities. Secondly, we design a Cross-Modal Adapter to augment the semantic richness of multimodal representations through cross-modal knowledge injection. Extensive domain experiments conducted on three datasets demonstrate that the proposed MIDG achieves superior performance.

This paper proposes SR-KI, a novel approach for integrating real-time and large-scale structured knowledge bases (KBs) into large language models (LLMs). SR-KI begins by encoding KBs into key-value pairs using a pretrained encoder, and injects them into LLMs' KV cache. Building on this representation, we employ a two-stage training paradigm: first locating a dedicated retrieval layer within the LLM, and then applying an attention-based loss at this layer to explicitly supervise attention toward relevant KB entries. Unlike traditional retrieval-augmented generation methods that rely heavily on the performance of external retrievers and multi-stage pipelines, SR-KI supports end-to-end inference by performing retrieval entirely within the model's latent space. This design enables efficient compression of injected knowledge and facilitates dynamic knowledge updates. Comprehensive experiments demonstrate that SR-KI enables the integration of up to 40K KBs into a 7B LLM on a single A100 40GB GPU, and achieves strong retrieval performance, maintaining over 98% Recall@10 on the best-performing task and exceeding 88% on average across all tasks. Task performance on question answering and KB ID generation also demonstrates that SR-KI maintains strong performance while achieving up to 99.75% compression of the injected KBs. Our code will be available at SR-KI.

Large Multimodal Models encode extensive factual knowledge in their pre-trained weights. However, its knowledge remains static and limited, unable to keep pace with real-world developments, which hinders continuous knowledge acquisition. Effective knowledge injection thus becomes critical, involving two goals: knowledge adaptation (injecting new knowledge) and knowledge retention (preserving old knowledge). Existing methods often struggle to learn new knowledge and suffer from catastrophic forgetting. To address this, we propose KORE, a synergistic method of KnOwledge-oRientEd augmentations and constraints for injecting new knowledge into large multimodal models while preserving old knowledge. Unlike general text or image data augmentation, KORE automatically converts individual knowledge items into structured and comprehensive knowledge to ensure that the model accurately learns new knowledge, enabling accurate adaptation. Meanwhile, KORE stores previous knowledge in the covariance matrix of LMM's linear layer activations and initializes the adapter by projecting the original weights into the matrix's null space, defining a fine-tuning direction that minimizes interference with previous knowledge, enabling powerful retention. Extensive experiments on various LMMs, including LLaVA-v1.5-7B, LLaVA-v1.5-13B, and Qwen2.5-VL-7B, show that KORE achieves superior new knowledge injection performance and effectively mitigates catastrophic forgetting.

Conventional continual pretraining (CPT) for large language model (LLM) domain adaptation often suffers from catastrophic forgetting and limited domain capacity. Existing strategies adopt layer expansion, introducing additional trainable parameters to accommodate new knowledge. However, the uniform expansion and updates still entangle general and domain learning, undermining its effectiveness. Our pilot studies reveal that LLMs exhibit functional specialization, where layers and units differentially encode general-critical capabilities, suggesting that parameter expansion and optimization should be function-aware. We then propose ADEPT, Adaptive Expansion and Dynamic Decoupled Tuning for continual pretraining, a two-stage framework for domain-adaptive CPT. ADEPT first performs General-Competence Guided Selective Layer Expansion, duplicating layers least critical for the general domain to increase representational capacity while minimizing interference with general knowledge. It then applies Adaptive Unit-Wise Decoupled Tuning, disentangling parameter units within expanded layers according to their general-domain importance and assigning asymmetric learning rates to balance knowledge injection and retention. Experiments on mathematical and medical benchmarks show that ADEPT outperforms full-parameter CPT by up to 5.76% on the general domain and 5.58% on the target domain with only 15% of parameters tuned and less than 50% training time. Ablation studies, theoretical analysis, and extended investigations further demonstrate the necessity of targeted expansion and decoupled optimization, providing new principles for efficient and robust domain-adaptive CPT. Our code is open-sourced at https://github.com/PuppyKnightUniversity/ADEPT

As Large Language Models (LLMs) are increasingly deployed in specialized domains with continuously evolving knowledge, the need for timely and precise knowledge injection has become essential.

Low-Rank Adaptation (LoRA) is widely adopted for downstream fine-tuning of foundation models due to its efficiency and zero additional inference cost. Many real-world applications require foundation models to specialize in several specific tasks simultaneously, motivating the need for efficient multi-task downstream adaptation. To address this need, existing studies have primarily explored two directions: Model Merging with LoRA, which shows advantages in training-free scenarios but still lags behind multi-task training in overall performance; and MoE-based LoRA approaches, which improve multi-task learning performance but introduce routers that hinder the mergeability of LoRA parameters and incur considerable inference overhead, thereby limiting real-world deployment practicality. To this end, we propose ThanoRA, a Task Heterogeneity-Aware Multi-Task Low-Rank Adaptation framework that enables effective, efficient and unified multi-task downstream adaptation without introducing additional structure. ThanoRA performs multi-task learning by tailoring subspace allocation at initialization and enforcing diversity preservation throughout training: it allocates varying dimensions to construct task-specific low-rank subspaces driven by inter-task heterogeneity, enabling fine-grained knowledge injection, while diversity-preserving regularization mitigates task interference and subspace collapse, thereby fully exploiting the low-rank capacity. Extensive experiments across multimodal and text-only benchmarks under varying multi-task mixtures demonstrate that ThanoRA consistently outperforms strong baselines, surpassing even separate task-specific fine-tuning, while introducing no additional structures or inference overhead. Our code will be publicly available at: https://github.com/LiangJian24/ThanoRA.

Depression is a growing concern gaining attention in both public discourse and AI research. While deep neural networks (DNNs) have been used for recognition, they still lack real-world effectiveness. Large language models (LLMs) show strong potential but require domain-specific fine-tuning and struggle with non-textual cues. Since depression is often expressed through vocal tone and behaviour rather than explicit text, relying on language alone is insufficient. Diagnostic accuracy also suffers without incorporating psychological expertise. To address these limitations, we present, to the best of our knowledge, the first application of LLMs to multimodal depression detection using the DAIC-WOZ dataset. We extract the audio features using the pre-trained model Wav2Vec, and mapped it to text-based LLMs for further processing. We also propose a novel strategy for incorporating psychological knowledge into LLMs to enhance diagnostic performance, specifically using a question and answer set to grant authorised knowledge to LLMs. Our approach yields a notable improvement in both Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) compared to a base score proposed by the related original paper. The codes are available at https://github.com/myxp-lyp/Depression-detection.git

Large language models (LLMs) often require vast amounts of text to effectively acquire new knowledge. While continuing pre-training on large corpora or employing retrieval-augmented generation (RAG) has proven successful, updating an LLM with only a few thousand or million tokens remains challenging. In this work, we investigate the task of injecting small, unstructured information into LLMs and its relation to the catastrophic forgetting phenomenon. We use a dataset of recent news -- ensuring no overlap with the model's pre-training data -- to evaluate the knowledge acquisition by probing the model with question-answer pairs related the learned information. Starting from a continued pre-training baseline, we explored different augmentation algorithms to generate synthetic data to improve the knowledge acquisition capabilities. Our experiments show that simply continuing pre-training on limited data yields modest improvements, whereas exposing the model to diverse textual variations significantly improves the learning of new facts -- particularly with methods that induce greater variability through diverse prompting. Furthermore, we shed light on the forgetting phenomenon in small-data regimes, illustrating the delicate balance between learning new content and retaining existing capabilities. We also confirm the sensitivity of RAG-based approaches for knowledge injection, which often lead to greater degradation on control datasets compared to parametric methods. Finally, we demonstrate that models can generate effective synthetic training data themselves, suggesting a pathway toward self-improving model updates. All code and generated data used in our experiments are publicly available, providing a resource for studying efficient knowledge injection in LLMs with limited data at https://github.com/hugoabonizio/knowledge-injection-methods.