Pre-trained large language models (PLLMs) like OpenAI ChatGPT and Google Gemini face challenges such as inaccurate factual recall, hallucinations, biases, and future data leakage for temporal Knowledge Graph (tKG) forecasting. To address these issues, we introduce sLA-tKGF (small-scale language assistant for tKG forecasting), which utilizes Retrieval-Augmented Generation (RAG) aided, custom-trained small-scale language models through a tabula rasa approach from scratch for effective tKG forecasting. Our framework constructs knowledge-infused prompts with relevant historical data from tKGs, web search results, and PLLMs-generated textual descriptions to understand historical entity relationships prior to the target time. It leverages these external knowledge-infused prompts for deeper understanding and reasoning of context-specific semantic and temporal information to zero-shot prompt small-scale language models for more accurate predictions of future events within tKGs. It reduces hallucinations and mitigates distributional shift challenges through comprehending changing trends over time. As a result, it enables more accurate and contextually grounded forecasts of future events while minimizing computational demands. Rigorous empirical studies demonstrate our framework robustness, scalability, and state-of-the-art (SOTA) performance on benchmark datasets with interpretable and trustworthy tKG forecasting.

Time series modeling is crucial for many applications, however, it faces challenges such as complex spatio-temporal dependencies and distribution shifts in learning from historical context to predict task-specific outcomes. To address these challenges, we propose a novel approach using an agentic Retrieval-Augmented Generation (RAG) framework for time series analysis. The framework leverages a hierarchical, multi-agent architecture where the master agent orchestrates specialized sub-agents and delegates the end-user request to the relevant sub-agent. The sub-agents utilize smaller, pre-trained language models (SLMs) customized for specific time series tasks through fine-tuning using instruction tuning and direct preference optimization, and retrieve relevant prompts from a shared repository of prompt pools containing distilled knowledge about historical patterns and trends to improve predictions on new data. Our proposed modular, multi-agent RAG approach offers flexibility and achieves state-of-the-art performance across major time series tasks by tackling complex challenges more effectively than task-specific customized methods across benchmark datasets.

Text embedding models are fundamental in natural language processing (NLP), serving as lowdimensional vector representations that capture semantic similarity between texts (Aggarwal & Zhai, 2012; Angelov, 2020). They are critical for tasks such as text classification, retrieval, question answering, and dialogue systems. Recent advancements in large language models (LLMs) have spurred interest in retrieval-augmented systems that integrate LLM reasoning with the efficiency of text embeddings. Two main research directions exist: enhancing performance in semantic textual similarity (STS) through supervised fine-tuning, normalization, and unsupervised contrastive learning; and addressing text retrieval through dual-encoder architectures and self-supervised pre-training (Li et al., 2023; Wang et al., 2022; Izacard & Grave, 2020; Ren et al., 2021;; Devlin, 2018; Vaswani, 2017). Recent studies have aimed to create unified text representation models through large-scale contrastive learning and prompt-based learning (Muennighoff, 2022), evaluated on benchmarks like the massive text embedding benchmark (MTEB) (Muennighoff et al., 2022), which assesses models across 56 datasets and seven categories.

Molecule design is a multifaceted approach that leverages computational methods and experiments to optimize molecular properties, fast-tracking new drug discoveries, innovative material development, and more efficient chemical processes. Recently, text-based molecule design has emerged, inspired by next-generation AI tasks analogous to foundational vision-language models. Our study explores the use of knowledge-augmented prompting of large language models (LLMs) for the zero-shot text-conditional de novo molecular generation task. Our approach uses task-specific instructions and a few demonstrations to address distributional shift challenges when constructing augmented prompts for querying LLMs to generate molecules consistent with technical descriptions. Our framework proves effective, outperforming state-of-the-art (SOTA) baseline models on benchmark datasets.

Legal Judgment Prediction (LJP) aims to automatically predict a law case's judgment results based on the text description of its facts. In practice, the confusing law articles (or charges) problem frequently occurs, reflecting that the law cases applicable to similar articles (or charges) tend to be misjudged. Although some recent works based on prior knowledge solve this issue well, they ignore that confusion also occurs between law articles with a high posterior semantic similarity due to the data imbalance problem instead of only between the prior highly similar ones, which is this work's further finding. This paper proposes an end-to-end model named \textit{D-LADAN} to solve the above challenges. On the one hand, D-LADAN constructs a graph among law articles based on their text definition and proposes a graph distillation operation (GDO) to distinguish the ones with a high prior semantic similarity. On the other hand, D-LADAN presents a novel momentum-updated memory mechanism to dynamically sense the posterior similarity between law articles (or charges) and a weighted GDO to adaptively capture the distinctions for revising the inductive bias caused by the data imbalance problem. We perform extensive experiments to demonstrate that D-LADAN significantly outperforms state-of-the-art methods in accuracy and robustness.

Procedural Content Generation via Machine Learning (PCGML) has enhanced game content creation, yet challenges in controllability and limited training data persist. This study addresses these issues by distilling a constructive PCG algorithm into a controllable PCGML model. We first generate a large amount of content with a constructive algorithm and label it using a Large Language Model (LLM). We use these synthetic labels to condition two PCGML models for content-specific generation, a diffusion model and the five-dollar model. This neural network distillation process ensures that the generation aligns with the original algorithm while introducing controllability through plain text. We define this text-conditioned PCGML as a Text-to-game-Map (T2M) task, offering an alternative to prevalent text-to-image multi-modal tasks. We compare our distilled models with the baseline constructive algorithm. Our analysis of the variety, accuracy, and quality of our generation demonstrates the efficacy of distilling constructive methods into controllable text-conditioned PCGML models.

The grammatical knowledge of language models (LMs) is often measured using a benchmark of linguistic minimal pairs, where LMs are presented with a pair of acceptable and unacceptable sentences and required to judge which is acceptable. The existing dominant approach, however, naively calculates and compares the probabilities of paired sentences using LMs. Additionally, large language models (LLMs) have yet to be thoroughly examined in this field. We thus investigate how to make the most of LLMs' grammatical knowledge to comprehensively evaluate it. Through extensive experiments of nine judgment methods in English and Chinese, we demonstrate that a probability readout method, in-template LP, and a prompting-based method, Yes/No probability computing, achieve particularly high performance, surpassing the conventional approach. Our analysis reveals their different strengths, e.g., Yes/No probability computing is robust against token-length bias, suggesting that they harness different aspects of LLMs' grammatical knowledge. Consequently, we recommend using diverse judgment methods to evaluate LLMs comprehensively.

Mixture-of-Experts (MoE) models are designed to enhance the efficiency of large language models (LLMs) without proportionally increasing the computational demands. However, their deployment on edge devices still faces significant challenges due to high on-demand loading overheads from managing sparsely activated experts. This paper introduces AdapMoE, an algorithm-system co-design framework for efficient MoE inference. AdapMoE features adaptive expert gating and management to reduce the on-demand loading overheads. We observe the heterogeneity of experts loading across layers and tokens, based on which we propose a sensitivity-based strategy to adjust the number of activated experts dynamically. Meanwhile, we also integrate advanced prefetching and cache management techniques to further reduce the loading latency. Through comprehensive evaluations on various platforms, we demonstrate AdapMoE consistently outperforms existing techniques, reducing the average number of activated experts by 25% and achieving a 1.35x speedup without accuracy degradation. Code is available at: https://github.com/PKU-SEC-Lab/AdapMoE.

Simultaneous speech translation (SST) takes streaming speech input and generates text translation on the fly. Existing methods either have high latency due to recomputation of input representations, or fall behind of offline ST in translation quality. In this paper, we propose FASST, a fast large language model based method for streaming speech translation. We propose blockwise-causal speech encoding and consistency mask, so that streaming speech input can be encoded incrementally without recomputation. Furthermore, we develop a two-stage training strategy to optimize FASST for simultaneous inference. We evaluate FASST and multiple strong prior models on MuST-C dataset. Experiment results show that FASST achieves the best quality-latency trade-off. It outperforms the previous best model by an average of 1.5 BLEU under the same latency for English to Spanish translation.

Prevailing large language models (LLMs) are capable of human responses simulation through its unprecedented content generation and reasoning abilities. However, it is not clear whether and how to leverage LLMs to simulate field experiments. In this paper, we propose and evaluate two prompting strategies: the observer mode that allows a direct prediction on main conclusions and the participant mode that simulates distributions of responses from participants. Using this approach, we examine fifteen well cited field experimental papers published in INFORMS and MISQ, finding encouraging alignments between simulated experimental results and the actual results in certain scenarios. We further identify topics of which LLMs underperform, including gender difference and social norms related research. Additionally, the automatic and standardized workflow proposed in this paper enables the possibility of a large-scale screening of more papers with field experiments. This paper pioneers the utilization of large language models (LLMs) for simulating field experiments, presenting a significant extension to previous work which focused solely on lab environments. By introducing two novel prompting strategies, observer and participant modes, we demonstrate the ability of LLMs to both predict outcomes and replicate participant responses within complex field settings. Our findings indicate a promising alignment with actual experimental results in certain scenarios, achieving a stimulation accuracy of 66% in observer mode. This study expands the scope of potential applications for LLMs and illustrates their utility in assisting researchers prior to engaging in expensive field experiments. Moreover, it sheds light on the boundaries of LLMs when used in simulating field experiments, serving as a cautionary note for researchers considering the integration of LLMs into their experimental toolkit.