A Watermark for Order-Agnostic Language Models

Statistical watermarking techniques are well-established for sequentially decoded language models (LMs). However, these techniques cannot be directly applied to order-agnostic LMs, as the tokens in order-agnostic LMs are not generated sequentially. In this work, we introduce Pattern-mark, a pattern-based watermarking framework specifically designed for order-agnostic LMs. We develop a Markov-chain-based watermark generator that produces watermark key sequences with high-frequency key patterns. Correspondingly, we propose a statistical pattern-based detection algorithm that recovers the key sequence during detection and conducts statistical tests based on the count of high-frequency patterns. Our extensive evaluations on order-agnostic LMs, such as ProteinMPNN and CMLM, demonstrate Pattern-mark's enhanced detection efficiency, generation quality, and robustness, positioning it as a superior watermarking technique for order-agnostic LMs.

A Practice of Post-Training on Llama-3 70B with Optimal Selection of Additional Language Mixture Ratio

Large Language Models (LLM) often needs to be Continual Pre-Trained (CPT) to obtain the unfamiliar language skill or adapt into new domains. The huge training cost of CPT often asks for cautious choice of key hyper-parameters such as the mixture ratio of extra language or domain corpus. However, there is no systematic study which bridge the gap between the optimal mixture ratio and the actual model performance, and the gap between experimental scaling law and the actual deployment in the full model size. In this paper, we perform CPT on Llama-3 8B and 70B to enhance its Chinese ability. We study the optimal correlation between the Additional Language Mixture Ratio (ALMR) and the Learning Rate (LR) on the 8B size which directly indicate the optimal experimental set up. By thorough choice of hyper-parameter, and subsequent fine-tuning, the model capability is improved not only on the Chinese-related benchmark, but also some specific domains including math, coding and emotional intelligence. We deploy the final 70B version of LLM on an real-life chat system which obtain satisfying performance.

Alleviating Hallucinations in Large Language Models with Scepticism Modeling

Hallucinations is a major challenge for large language models (LLMs), prevents adoption in diverse fields. Uncertainty estimation could be used for alleviating the damages of hallucinations. The skeptical emotion of human could be useful for enhancing the ability of self estimation. Inspirited by this observation, we proposed a new approach called Skepticism Modeling (SM). This approach is formalized by combining the information of token and logits for self estimation. We construct the doubt emotion aware data, perform continual pre-training, and then fine-tune the LLMs, improve their ability of self estimation. Experimental results demonstrate this new approach effectively enhances a model's ability to estimate their uncertainty, and validate its generalization ability of other tasks by out-of-domain experiments.

Detection and Multi-Parameter Estimation for NLOS Targets: An IRS-assisted Framework

Intelligent reflecting surface (IRS) has the potential to enhance sensing performance, due to its capability of reshaping the echo signals. Different from the existing literature, which has commonly focused on IRS beamforming optimization, in this paper, we pay special attention to designing effective signal processing approaches to extract sensing information from IRS-reshaped echo signals. To this end, we investigate an IRS-assisted non-line-of-sight (NLOS) target detection and multi-parameter estimation problem in orthogonal frequency division multiplexing (OFDM) systems. To address this problem, we first propose a novel detection and direction estimation framework, including a low-overhead hierarchical codebook that allows the IRS to generate three-dimensional beams with adjustable beam direction and width, a delay spectrum peak-based beam training scheme for detection and direction estimation, and a beam refinement scheme for further enhancing the accuracy of the direction estimation. Then, we propose a target range and velocity estimation scheme by extracting the delay-Doppler information from the IRS-reshaped echo signals. Numerical results demonstrate that the proposed schemes can achieve 99.7% target detection rate, a 10^{-3}-rad level direction estimation accuracy, and a 10^{-6}-m/10^{-5}-m/s level range/velocity estimation accuracy.

Rethinking the fundamental performance limits of integrated sensing and communication systems

Integrated sensing and communication (ISAC) has been recognized as a key enabler and feature of future wireless networks. In the existing works analyzing the performances of ISAC, discrete-time systems were commonly assumed, which, however, overlooked the impacts of temporal, spectral, and spatial properties. To address this issue, we establish a unified information model for the band-limited continuous-time ISAC systems. In the established information model, we employ a novel sensing performance metric, called the sensing mutual information (SMI). Through analysis, we show how the SMI can be utilized as a bridge between the mutual information domain and the mean squared error (MSE) domain. In addition, we illustrate the communication mutual information (CMI)-SMI and CMI-MSE regions to identify the performance bounds of ISAC systems in practical settings and reveal the trade-off between communication and sensing performances. Moreover, via analysis and numerical results, we provide two valuable insights into the design of novel ISAC-enabled systems: i) communication prefers the waveforms of random amplitude, sensing prefers the waveforms of constant amplitude, both communication and sensing favor the waveforms of low correlations with random phases; ii) There exists a linear positive proportional relationship between the allocated time-frequency resource and the achieved communication rate/sensing MSE.

OPTune: Efficient Online Preference Tuning

Reinforcement learning with human feedback~(RLHF) is critical for aligning Large Language Models (LLMs) with human preference. Compared to the widely studied offline version of RLHF, \emph{e.g.} direct preference optimization (DPO), recent works have shown that the online variants achieve even better alignment. However, online alignment requires on-the-fly generation of new training data, which is costly, hard to parallelize, and suffers from varying quality and utility. In this paper, we propose a more efficient data exploration strategy for online preference tuning (OPTune), which does not rely on human-curated or pre-collected teacher responses but dynamically samples informative responses for on-policy preference alignment. During data generation, OPTune only selects prompts whose (re)generated responses can potentially provide more informative and higher-quality training signals than the existing responses. In the training objective, OPTune reweights each generated response (pair) by its utility in improving the alignment so that learning can be focused on the most helpful samples. Throughout our evaluations, OPTune'd LLMs maintain the instruction-following benefits provided by standard preference tuning whilst enjoying 1.27-1.56x faster training speed due to the efficient data exploration strategy.

TimeCMA: Towards LLM-Empowered Time Series Forecasting via Cross-Modality Alignment

The widespread adoption of scalable mobile sensing has led to large amounts of time series data for real-world applications. A fundamental application is multivariate time series forecasting (MTSF), which aims to predict future time series values based on historical observations. Existing MTSF methods suffer from limited parameterization and small-scale training data. Recently, Large language models (LLMs) have been introduced in time series, which achieve promising forecasting performance but incur heavy computational costs. To solve these challenges, we propose TimeCMA, an LLM-empowered framework for time series forecasting with cross-modality alignment. We design a dual-modality encoding module with two branches, where the time series encoding branch extracts relatively low-quality yet pure embeddings of time series through an inverted Transformer. In addition, the LLM-empowered encoding branch wraps the same time series as prompts to obtain high-quality yet entangled prompt embeddings via a Pre-trained LLM. Then, we design a cross-modality alignment module to retrieve high-quality and pure time series embeddings from the prompt embeddings. Moreover, we develop a time series forecasting module to decode the aligned embeddings while capturing dependencies among multiple variables for forecasting. Notably, we tailor the prompt to encode sufficient temporal information into a last token and design the last token embedding storage to reduce computational costs. Extensive experiments on real data offer insight into the accuracy and efficiency of the proposed framework.

SQL-to-Schema Enhances Schema Linking in Text-to-SQL

In sophisticated existing Text-to-SQL methods exhibit errors in various proportions, including schema-linking errors (incorrect columns, tables, or extra columns), join errors, nested errors, and group-by errors. Consequently, there is a critical need to filter out unnecessary tables and columns, directing the language models attention to relevant tables and columns with schema-linking, to reduce errors during SQL generation. Previous approaches have involved sorting tables and columns based on their relevance to the question, selecting the top-ranked ones for sorting, or directly identifying the necessary tables and columns for SQL generation. However, these methods face challenges such as lengthy model training times, high consumption of expensive GPT-4 tokens in few-shot prompts, or suboptimal performance in schema linking. Therefore, we propose an inventive schema linking method in two steps: Firstly, generate an initial SQL query by utilizing the complete database schema. Subsequently, extract tables and columns from the initial SQL query to create a concise schema. Using CodeLlama-34B, when comparing the schemas obtained by mainstream methods with ours for SQL generation, our schema performs optimally. Leveraging GPT4, our SQL generation method achieved results that are comparable to mainstream Text-to-SQL methods on the Spider dataset.

LightTR: A Lightweight Framework for Federated Trajectory Recovery

With the proliferation of GPS-equipped edge devices, huge trajectory data is generated and accumulated in various domains, motivating a variety of urban applications. Due to the limited acquisition capabilities of edge devices, a lot of trajectories are recorded at a low sampling rate, which may lead to the effectiveness drop of urban applications. We aim to recover a high-sampled trajectory based on the low-sampled trajectory in free space, i.e., without road network information, to enhance the usability of trajectory data and support urban applications more effectively. Recent proposals targeting trajectory recovery often assume that trajectories are available at a central location, which fail to handle the decentralized trajectories and hurt privacy. To bridge the gap between decentralized training and trajectory recovery, we propose a lightweight framework, LightTR, for federated trajectory recovery based on a client-server architecture, while keeping the data decentralized and private in each client/platform center (e.g., each data center of a company). Specifically, considering the limited processing capabilities of edge devices, LightTR encompasses a light local trajectory embedding module that offers improved computational efficiency without compromising its feature extraction capabilities. LightTR also features a meta-knowledge enhanced local-global training scheme to reduce communication costs between the server and clients and thus further offer efficiency improvement. Extensive experiments demonstrate the effectiveness and efficiency of the proposed framework.

MuseumMaker: Continual Style Customization without Catastrophic Forgetting

Pre-trained large text-to-image (T2I) models with an appropriate text prompt has attracted growing interests in customized images generation field. However, catastrophic forgetting issue make it hard to continually synthesize new user-provided styles while retaining the satisfying results amongst learned styles. In this paper, we propose MuseumMaker, a method that enables the synthesis of images by following a set of customized styles in a never-end manner, and gradually accumulate these creative artistic works as a Museum. When facing with a new customization style, we develop a style distillation loss module to extract and learn the styles of the training data for new image generation. It can minimize the learning biases caused by content of new training images, and address the catastrophic overfitting issue induced by few-shot images. To deal with catastrophic forgetting amongst past learned styles, we devise a dual regularization for shared-LoRA module to optimize the direction of model update, which could regularize the diffusion model from both weight and feature aspects, respectively. Meanwhile, to further preserve historical knowledge from past styles and address the limited representability of LoRA, we consider a task-wise token learning module where a unique token embedding is learned to denote a new style. As any new user-provided style come, our MuseumMaker can capture the nuances of the new styles while maintaining the details of learned styles. Experimental results on diverse style datasets validate the effectiveness of our proposed MuseumMaker method, showcasing its robustness and versatility across various scenarios.