Despite significant advancements in large language models (LLMs), the rapid and frequent integration of small-scale experiences, such as interactions with sur- rounding objects, remains a substantial challenge. Two critical factors in assimilating these experiences are (1) **Efficacy**: the ability to accurately remember recent events; (2) **Retention**: the capacity to recall long-past experiences. Current methods either embed experiences within model parameters using continual learning, model editing, or knowledge distillation techniques, which often struggle with rapid updates and complex interactions, or rely on external storage to achieve long-term retention, thereby increasing storage requirements. In this paper, we propose **SELF-PARAM** (Self-Updatable Large Language Models with Parameter Integration). SELF-PARAM requires no extra parameters while ensuring near-optimal efficacy and long-term retention. Our method employs a training objective that minimizes the Kullback-Leibler (KL) divergence between the predictions of an original model (with access to contextual information) and a target model (without such access). By generating diverse question-answer pairs related to the knowledge and minimizing the KL divergence across this dataset, we update the target model to internalize the knowledge seamlessly within its parameters. Evaluations on question-answering and conversational recommendation tasks demonstrate that SELF-PARAM significantly outperforms existing methods, even when accounting for non-zero storage requirements. This advancement paves the way for more efficient and scalable integration of experiences in large language models by embedding knowledge directly into model parameters.

Large language models show impressive abilities in memorizing world knowledge, which leads to concerns regarding memorization of private information, toxic or sensitive knowledge, and copyrighted content. We introduce the problem of Large Scale Knowledge Washing, focusing on unlearning an extensive amount of factual knowledge. Previous unlearning methods usually define the reverse loss and update the model via backpropagation, which may affect the model's fluency and reasoning ability or even destroy the model due to extensive training with the reverse loss. Existing works introduce additional data from downstream tasks to prevent the model from losing capabilities, which requires downstream task awareness. Controlling the tradeoff of unlearning existing knowledge while maintaining existing capabilities is also challenging. To this end, we propose LaW (Large Scale Washing), where we update the MLP layers in decoder-only large language models to perform knowledge washing, as inspired by model editing methods. We derive a new objective with the knowledge to be unlearned to update the weights of certain MLP layers. Experimental results demonstrate the effectiveness of LaW in forgetting target knowledge while maximally maintaining reasoning ability. The code will be open-sourced.

Equipping large language models (LLMs) with latent-space memory has attracted increasing attention as they can extend the context window of existing language models. However, retaining information from the distant past remains a challenge. For example, MemoryLLM (Wang et al., 2024a), as a representative work with latent-space memory, compresses past information into hidden states across all layers, forming a memory pool of 1B parameters. While effective for sequence lengths up to 16k tokens, it struggles to retain knowledge beyond 20k tokens. In this work, we address this limitation by introducing M+, a memory-augmented model based on MemoryLLM that significantly enhances long-term information retention. M+ integrates a long-term memory mechanism with a co-trained retriever, dynamically retrieving relevant information during text generation. We evaluate M+ on diverse benchmarks, including long-context understanding and knowledge retention tasks. Experimental results show that M+ significantly outperforms MemoryLLM and recent strong baselines, extending knowledge retention from under 20k to over 160k tokens with similar GPU memory overhead.

Conversational recommender systems engage users in dialogues to refine their needs and provide more personalized suggestions. Although textual information suffices for many domains, visually driven categories such as fashion or home decor potentially require detailed visual information related to color, style, or design. To address this challenge, we propose LaViC (Large Vision-Language Conversational Recommendation Framework), a novel approach that integrates compact image representations into dialogue-based recommendation systems. LaViC leverages a large vision-language model in a two-stage process: (1) visual knowledge self-distillation, which condenses product images from thousands of tokens into a small set of visual tokens in a self-distillation manner, significantly reducing computational overhead, and (2) recommendation fine-tuning, which enables the model to incorporate both dialogue context and distilled visual tokens, providing a unified mechanism for capturing textual and visual features. To support rigorous evaluation of visually-aware conversational recommendation, we construct a new dataset by aligning Reddit conversations with Amazon product listings across multiple visually oriented categories (e.g., fashion, beauty, and home). This dataset covers realistic user queries and product appearances in domains where visual details are crucial. Extensive experiments demonstrate that LaViC significantly outperforms text-only conversational recommendation methods and open-source vision-language baselines. Moreover, LaViC achieves competitive or superior accuracy compared to prominent proprietary baselines (e.g., GPT-3.5-turbo, GPT-4o-mini, and GPT-4o), demonstrating the necessity of explicitly using visual data for capturing product attributes and showing the effectiveness of our vision-language integration. Our code and dataset are available at https://github.com/jeon185/LaViC.

Recent research on explainable recommendation generally frames the task as a standard text generation problem, and evaluates models simply based on the textual similarity between the predicted and ground-truth explanations. However, this approach fails to consider one crucial aspect of the systems: whether their outputs accurately reflect the users' (post-purchase) sentiments, i.e., whether and why they would like and/or dislike the recommended items. To shed light on this issue, we introduce new datasets and evaluation methods that focus on the users' sentiments. Specifically, we construct the datasets by explicitly extracting users' positive and negative opinions from their post-purchase reviews using an LLM, and propose to evaluate systems based on whether the generated explanations 1) align well with the users' sentiments, and 2) accurately identify both positive and negative opinions of users on the target items. We benchmark several recent models on our datasets and demonstrate that achieving strong performance on existing metrics does not ensure that the generated explanations align well with the users' sentiments. Lastly, we find that existing models can provide more sentiment-aware explanations when the users' (predicted) ratings for the target items are directly fed into the models as input. The datasets and benchmark implementation are available at: https://github.com/jchanxtarov/sent_xrec.

Understanding and accurately explaining compatibility relationships between fashion items is a challenging problem in the burgeoning domain of AI-driven outfit recommendations. Present models, while making strides in this area, still occasionally fall short, offering explanations that can be elementary and repetitive. This work aims to address these shortcomings by introducing the Pair Fashion Explanation (PFE) dataset, a unique resource that has been curated to illuminate these compatibility relationships. Furthermore, we propose an innovative two stage pipeline model that leverages this dataset. This fine-tuning allows the model to generate explanations that convey the compatibility relationships between items. Our experiments showcase the model's potential in crafting descriptions that are knowledgeable, aligned with ground-truth matching correlations, and that produce understandable and informative descriptions, as assessed by both automatic metrics and human evaluation. Our code and data are released at https://github.com/wangyu-ustc/PairFashionExplanation.

The vast applications of deep generative models are anchored in three core capabilities---*generating* new instances, *reconstructing* inputs, and learning compact *representations*---across various data types, such as discrete text/protein sequences and continuous images. Existing model families, like variational autoencoders (VAEs), generative adversarial networks (GANs), autoregressive models, and (latent) diffusion models, generally excel in specific capabilities and data types but fall short in others. We introduce *Generalized* ***E****ncoding*-***D****ecoding ****D****iffusion ****P****robabilistic ****M****odels* (EDDPMs) which integrate the core capabilities for broad applicability and enhanced performance. EDDPMs generalize the Gaussian noising-denoising in standard diffusion by introducing parameterized encoding-decoding. Crucially, EDDPMs are compatible with the well-established diffusion model objective and training recipes, allowing effective learning of the encoder-decoder parameters *jointly* with diffusion. By choosing appropriate encoder/decoder (e.g., large language models), EDDPMs naturally apply to different data types. Extensive experiments on text, proteins, and images demonstrate the flexibility to handle diverse data and tasks and the strong improvement over various existing models. Code is available at https://github.com/guangyliu/EDDPM .

Existing Large Language Models (LLMs) usually remain static after deployment, which might make it hard to inject new knowledge into the model. We aim to build models containing a considerable portion of self-updatable parameters, enabling the model to integrate new knowledge effectively and efficiently. To this end, we introduce MEMORYLLM, a model that comprises a transformer and a fixed-size memory pool within the latent space of the transformer. MEMORYLLM can self-update with text knowledge and memorize the knowledge injected earlier. Our evaluations demonstrate the ability of MEMORYLLM to effectively incorporate new knowledge, as evidenced by its performance on model editing benchmarks. Meanwhile, the model exhibits long-term information retention capacity, which is validated through our custom-designed evaluations and long-context benchmarks. MEMORYLLM also shows operational integrity without any sign of performance degradation even after nearly a million memory updates. Our code and model are open-sourced at https://github.com/wangyu-ustc/MemoryLLM.

Outlier detection is crucial for preventing financial fraud, network intrusions, and device failures. Users often expect systems to automatically summarize and interpret outlier detection results to reduce human effort and convert outliers into actionable insights. However, existing methods fail to effectively assist users in identifying the root causes of outliers, as they only pinpoint data attributes without considering outliers in the same subspace may have different causes. To fill this gap, we propose STAIR, which learns concise and human-understandable rules to summarize and explain outlier detection results with finer granularity. These rules consider both attributes and associated values. STAIR employs an interpretation-aware optimization objective to generate a small number of rules with minimal complexity for strong interpretability. The learning algorithm of STAIR produces a rule set by iteratively splitting the large rules and is optimal in maximizing this objective in each iteration. Moreover, to effectively handle high dimensional, highly complex data sets that are hard to summarize with simple rules, we propose a localized STAIR approach, called L-STAIR. Taking data locality into consideration, it simultaneously partitions data and learns a set of localized rules for each partition. Our experimental study on many outlier benchmark datasets shows that STAIR significantly reduces the complexity of the rules required to summarize the outlier detection results, thus more amenable for humans to understand and evaluate.

Curated datasets for healthcare are often limited due to the need of human annotations from experts.In this paper, we present MEDEVAL, a multi-level, multi-task, and multi-domain medical benchmark to facilitate the development of language models for healthcare.MEDEVAL is comprehensive and consists of data from several healthcare systems and spans 35 human body regions from 8 examination modalities.With 22,779 collected sentences and 21,228 reports, we provide expert annotations at multiple levels, offering a granular potential usage of the data and supporting a wide range of tasks.Moreover, we systematically evaluated 10 generic and domain-specific language models under zero-shot and finetuning settings, from domain-adapted baselines in healthcare to general-purposed state-of-the-art large language models (e.g., ChatGPT).Our evaluations reveal varying effectiveness of the two categories of language models across different tasks, from which we notice the importance of instruction tuning for few-shot usage of large language models.Our investigation paves the way toward benchmarking language models for healthcare and provides valuable insights into the strengths and limitations of adopting large language models in medical domains, informing their practical applications and future advancements 1 . Chest

Recent advances in foundation models present new opportunities for interpretable visual recognition – one can first query Large Language Models (LLMs) to obtain a set of attributes that describe each class, then apply vision-language models to classify images via these attributes. Pioneering work shows that querying thousands of attributes can achieve performance competitive with image features. However, our further investigation on 8 datasets reveals that LLM-generated attributes in a large quantity perform almost the same as random words. This surprising finding suggests that significant noise may be present in these attributes. We hypothesize that there exist subsets of attributes that can maintain the classification performance with much smaller sizes, and propose a novel learning-to-search method to discover those concise sets of attributes. As a result, on the CUB dataset, our method achieves performance close to that of massive LLM-generated attributes (e.g., 10k attributes for CUB), yet using only 32 attributes in total to distinguish 200 bird species. Furthermore, our new paradigm demonstrates several additional benefits: higher interpretability and interactivity for humans, and the ability to summarize knowledge for a recognition task.

Outlier detection is critical in real world. Due to the existence of many outlier detection techniques which often return different results for the same data set, the users have to address the problem of determining which among these techniques is the best suited for their task and tune its parameters. This is particularly challenging in the unsupervised setting, where no labels are available for cross-validation needed for such method and parameter optimization. In this work, we propose AutoOD which uses the existing unsupervised detection techniques to automatically produce high quality outliers without any human tuning. AutoOD's fundamentally new strategy unifies the merits of unsupervised outlier detection and supervised classification within one integrated solution. It automatically tests a diverse set of unsupervised outlier detectors on a target data set, extracts useful signals from their combined detection results to reliably capture key differences between outliers and inliers. It then uses these signals to produce a "custom outlier classifier" to classify outliers, with its accuracy comparable to supervised outlier classification models trained with ground truth labels - without having access to the much needed labels. On a diverse set of benchmark outlier detection datasets, AutoOD consistently outperforms the best unsupervised outlier detector selected from hundreds of detectors. It also outperforms other tuning-free approaches from 12 to 97 points (out of 100) in the F-1 score.

Machine learning algorithms typically assume that training and test examples are drawn from the same distribution. However, distribution shift is a common problem in real-world applications and can cause models to perform dramatically worse at test time. In this paper, we specifically consider the problems of subpopulation shifts (e.g., imbalanced data) and domain shifts. While prior works often seek to explicitly regularize internal representations or predictors of the model to be domain invariant, we instead aim to learn invariant predictors without restricting the model’s internal representations or predictors. This leads to a simple mixup-based technique which learns invariant predictors via selective augmentation called LISA. LISA selectively interpolates samples either with the same labels but different domains or with the same domain but different labels. Empirically, we study the effectiveness of LISA on nine benchmarks ranging from subpopulation shifts to domain shifts, and we find that LISA consistently outperforms other state-of-the-art methods and leads to more invariant predictors. We further analyze a linear setting and theoretically show how LISA leads to a smaller worst-group error.

To benefit the learning of a new task, meta-learning has been proposed to transfer a well-generalized meta-model learned from various meta-training tasks. Existing meta-learning algorithms randomly sample meta-training tasks with a uniform probability, under the assumption that tasks are of equal importance. However, it is likely that tasks are detrimental with noise or imbalanced given a limited number of meta-training tasks. To prevent the meta-model from being corrupted by such detrimental tasks or dominated by tasks in the majority, in this paper, we propose an adaptive task scheduler (ATS) for the meta-training process. In ATS, for the first time, we design a neural scheduler to decide which meta-training tasks to use next by predicting the probability being sampled for each candidate task, and train the scheduler to optimize the generalization capacity of the meta-model to unseen tasks. We identify two meta-model-related factors as the input of the neural scheduler, which characterize the difficulty of a candidate task to the meta-model. Theoretically, we show that a scheduler taking the two factors into account improves the meta-training loss and also the optimization landscape. Under the setting of meta-learning with noise and limited budgets, ATS improves the performance on both miniImageNet and a real-world drug discovery benchmark by up to 13% and 18%, respectively, compared to state-of-the-art task schedulers.