Dial-insight: Fine-tuning Large Language Models with High-Quality Domain-Specific Data Preventing Capability Collapse

The efficacy of large language models (LLMs) is heavily dependent on the quality of the underlying data, particularly within specialized domains. A common challenge when fine-tuning LLMs for domain-specific applications is the potential degradation of the model's generalization capabilities. To address these issues, we propose a two-stage approach for the construction of production prompts designed to yield high-quality data. This method involves the generation of a diverse array of prompts that encompass a broad spectrum of tasks and exhibit a rich variety of expressions. Furthermore, we introduce a cost-effective, multi-dimensional quality assessment framework to ensure the integrity of the generated labeling data. Utilizing a dataset comprised of service provider and customer interactions from the real estate sector, we demonstrate a positive correlation between data quality and model performance. Notably, our findings indicate that the domain-specific proficiency of general LLMs can be enhanced through fine-tuning with data produced via our proposed method, without compromising their overall generalization abilities, even when exclusively domain-specific data is employed for fine-tuning.

A Physics-driven GraphSAGE Method for Physical Process Simulations Described by Partial Differential Equations

Physics-informed neural networks (PINNs) have successfully addressed various computational physics problems based on partial differential equations (PDEs). However, while tackling issues related to irregularities like singularities and oscillations, trained solutions usually suffer low accuracy. In addition, most current works only offer the trained solution for predetermined input parameters. If any change occurs in input parameters, transfer learning or retraining is required, and traditional numerical techniques also need an independent simulation. In this work, a physics-driven GraphSAGE approach (PD-GraphSAGE) based on the Galerkin method and piecewise polynomial nodal basis functions is presented to solve computational problems governed by irregular PDEs and to develop parametric PDE surrogate models. This approach employs graph representations of physical domains, thereby reducing the demands for evaluated points due to local refinement. A distance-related edge feature and a feature mapping strategy are devised to help training and convergence for singularity and oscillation situations, respectively. The merits of the proposed method are demonstrated through a couple of cases. Moreover, the robust PDE surrogate model for heat conduction problems parameterized by the Gaussian random field source is successfully established, which not only provides the solution accurately but is several times faster than the finite element method in our experiments.

CasCast: Skillful High-resolution Precipitation Nowcasting via Cascaded Modelling

Precipitation nowcasting based on radar data plays a crucial role in extreme weather prediction and has broad implications for disaster management. Despite progresses have been made based on deep learning, two key challenges of precipitation nowcasting are not well-solved: (i) the modeling of complex precipitation system evolutions with different scales, and (ii) accurate forecasts for extreme precipitation. In this work, we propose CasCast, a cascaded framework composed of a deterministic and a probabilistic part to decouple the predictions for mesoscale precipitation distributions and small-scale patterns. Then, we explore training the cascaded framework at the high resolution and conducting the probabilistic modeling in a low dimensional latent space with a frame-wise-guided diffusion transformer for enhancing the optimization of extreme events while reducing computational costs. Extensive experiments on three benchmark radar precipitation datasets show that CasCast achieves competitive performance. Especially, CasCast significantly surpasses the baseline (up to +91.8%) for regional extreme-precipitation nowcasting.

From LLM to Conversational Agent: A Memory Enhanced Architecture with Fine-Tuning of Large Language Models

This paper introduces RAISE (Reasoning and Acting through Scratchpad and Examples), an advanced architecture enhancing the integration of Large Language Models (LLMs) like GPT-4 into conversational agents. RAISE, an enhancement of the ReAct framework, incorporates a dual-component memory system, mirroring human short-term and long-term memory, to maintain context and continuity in conversations. It entails a comprehensive agent construction scenario, including phases like Conversation Selection, Scene Extraction, CoT Completion, and Scene Augmentation, leading to the LLMs Training phase. This approach appears to enhance agent controllability and adaptability in complex, multi-turn dialogues. Our preliminary evaluations in a real estate sales context suggest that RAISE has some advantages over traditional agents, indicating its potential for broader applications. This work contributes to the AI field by providing a robust framework for developing more context-aware and versatile conversational agents.

DUMA: a Dual-Mind Conversational Agent with Fast and Slow Thinking

Inspired by the dual-process theory of human cognition, we introduce DUMA, a novel conversational agent framework that embodies a dual-mind mechanism through the utilization of two generative Large Language Models (LLMs) dedicated to fast and slow thinking respectively. The fast thinking model serves as the primary interface for external interactions and initial response generation, evaluating the necessity for engaging the slow thinking model based on the complexity of the complete response. When invoked, the slow thinking model takes over the conversation, engaging in meticulous planning, reasoning, and tool utilization to provide a well-analyzed response. This dual-mind configuration allows for a seamless transition between intuitive responses and deliberate problem-solving processes based on the situation. We have constructed a conversational agent to handle online inquiries in the real estate industry. The experiment proves that our method balances effectiveness and efficiency, and has a significant improvement compared to the baseline.

A Survey on Hyperspectral Image Restoration: From the View of Low-Rank Tensor Approximation

The ability of capturing fine spectral discriminative information enables hyperspectral images (HSIs) to observe, detect and identify objects with subtle spectral discrepancy. However, the captured HSIs may not represent true distribution of ground objects and the received reflectance at imaging instruments may be degraded, owing to environmental disturbances, atmospheric effects and sensors' hardware limitations. These degradations include but are not limited to: complex noise (i.e., Gaussian noise, impulse noise, sparse stripes, and their mixtures), heavy stripes, deadlines, cloud and shadow occlusion, blurring and spatial-resolution degradation and spectral absorption, etc. These degradations dramatically reduce the quality and usefulness of HSIs. Low-rank tensor approximation (LRTA) is such an emerging technique, having gained much attention in HSI restoration community, with ever-growing theoretical foundation and pivotal technological innovation. Compared to low-rank matrix approximation (LRMA), LRTA is capable of characterizing more complex intrinsic structure of high-order data and owns more efficient learning abilities, being established to address convex and non-convex inverse optimization problems induced by HSI restoration. This survey mainly attempts to present a sophisticated, cutting-edge, and comprehensive technical survey of LRTA toward HSI restoration, specifically focusing on the following six topics: Denoising, Destriping, Inpainting, Deblurring, Super--resolution and Fusion. The theoretical development and variants of LRTA techniques are also elaborated. For each topic, the state-of-the-art restoration methods are compared by assessing their performance both quantitatively and visually. Open issues and challenges are also presented, including model formulation, algorithm design, prior exploration and application concerning the interpretation requirements.

Detecting Textual Adversarial Examples Based on Distributional Characteristics of Data Representations

Although deep neural networks have achieved state-of-the-art performance in various machine learning tasks, adversarial examples, constructed by adding small non-random perturbations to correctly classified inputs, successfully fool highly expressive deep classifiers into incorrect predictions. Approaches to adversarial attacks in natural language tasks have boomed in the last five years using character-level, word-level, phrase-level, or sentence-level textual perturbations. While there is some work in NLP on defending against such attacks through proactive methods, like adversarial training, there is to our knowledge no effective general reactive approaches to defence via detection of textual adversarial examples such as is found in the image processing literature. In this paper, we propose two new reactive methods for NLP to fill this gap, which unlike the few limited application baselines from NLP are based entirely on distribution characteristics of learned representations: we adapt one from the image processing literature (Local Intrinsic Dimensionality (LID)), and propose a novel one (MultiDistance Representation Ensemble Method (MDRE)). Adapted LID and MDRE obtain state-of-the-art results on character-level, word-level, and phrase-level attacks on the IMDB dataset as well as on the later two with respect to the MultiNLI dataset. For future research, we publish our code.

Metasurface-Enabled On-Chip Multiplexed Diffractive Neural Networks in the Visible

Replacing electrons with photons is a compelling route towards light-speed, highly parallel, and low-power artificial intelligence computing. Recently, all-optical diffractive neural deep neural networks have been demonstrated. However, the existing architectures often comprise bulky components and, most critically, they cannot mimic the human brain for multitasking. Here, we demonstrate a multi-skilled diffractive neural network based on a metasurface device, which can perform on-chip multi-channel sensing and multitasking at the speed of light in the visible. The metasurface is integrated with a complementary metal oxide semiconductor imaging sensor. Polarization multiplexing scheme of the subwavelength nanostructures are applied to construct a multi-channel classifier framework for simultaneous recognition of digital and fashionable items. The areal density of the artificial neurons can reach up to 6.25x106/mm2 multiplied by the number of channels. Our platform provides an integrated solution with all-optical on-chip sensing and computing for applications in machine vision, autonomous driving, and precision medicine.

Fine-Grained Age Estimation in the wild with Attention LSTM Networks

Age estimation from a single face image has been an essential task in the field of human-computer interaction and computer vision which has a wide range of practical application value. Concerning the problem that accuracy of age estimation of face images in the wild are relatively low for existing methods, where they take into account only the whole features of face image while neglecting the fine-grained features of age-sensitive area, we propose a method based on Attention LSTM network for Fine-Grained age estimation in the wild based on the idea of Fine-Grained categories and visual attention mechanism. This method combines ResNets or RoR models with LSTM unit to construct AL-ResNets or AL-RoR networks to extract age-sensitive local regions, which effectively improves age estimation accuracy. Firstly, ResNets or RoR model pre-trained on ImageNet dataset is selected as the basic model, which is then fine-tuned on the IMDB-WIKI-101 dataset for age estimation. Then, we fine-tune ResNets or RoR on the target age datasets to extract the global features of face images. To extract the local characteristics of age-sensitive areas, the LSTM unit is then presented to obtain the coordinates of the age-sensitive region automatically. Finally, the age group classification experiment is conducted directly on the Adience dataset, and age-regression experiments are performed by the Deep EXpectation algorithm (DEX) on MORPH Album 2, FG-NET and LAP datasets. By combining the global and local features, we got our final prediction results. Our experiments illustrate the effectiveness of AL-ResNets or AL-RoR for age estimation in the wild, where it achieves new state-of-the-art performance than all other CNN methods on the Adience, MORPH Album 2, FG-NET and LAP datasets.

Condition directed Multi-domain Adversarial Learning for Loop Closure Detection

Loop closure detection (LCD) is the key module in appearance based simultaneously localization and mapping (SLAM). However, in the real life, the appearance of visual inputs are usually affected by the illumination changes and texture changes under different weather conditions. Traditional methods in LCD usually rely on handcraft features, however, such methods are unable to capture the common descriptions under different weather conditions, such as rainy, foggy and sunny. Furthermore, traditional handcraft features could not capture the highly level understanding for the local scenes. In this paper, we proposed a novel condition directed multi-domain adversarial learning method, where we use the weather condition as the direction for feature inference. Based on the generative adversarial networks (GANs) and a classification networks, the proposed method could extract the high-level weather-invariant features directly from the raw data. The only labels required here are the weather condition of each visual input. Experiments are conducted in the GTAV game simulator, which could generated lifelike outdoor scenes under different weather conditions. The performance of LCD results shows that our method outperforms the state-of-arts significantly.