Large language models like ChatGPT have shown substantial progress in natural language understanding and generation, proving valuable across various disciplines, including the medical field. Despite advancements, challenges persist due to the complexity and diversity inherent in medical tasks which often require multi-task learning capabilities. Previous approaches, although beneficial, fall short in real-world applications because they necessitate task-specific annotations at inference time, limiting broader generalization. This paper introduces MING-MOE, a novel Mixture-of-Expert~(MOE)-based medical large language model designed to manage diverse and complex medical tasks without requiring task-specific annotations, thus enhancing its usability across extensive datasets. MING-MOE employs a Mixture of Low-Rank Adaptation (MoLoRA) technique, allowing for efficient parameter usage by maintaining base model parameters static while adapting through a minimal set of trainable parameters. We demonstrate that MING-MOE achieves state-of-the-art (SOTA) performance on over 20 medical tasks, illustrating a significant improvement over existing models. This approach not only extends the capabilities of medical language models but also improves inference efficiency.
Graph Neural Network has been proved to be effective for fraud detection for its capability to encode node interaction and aggregate features in a holistic view. Recently, Transformer network with great sequence encoding ability, has also outperformed other GNN-based methods in literatures. However, both GNN-based and Transformer-based networks only encode one perspective of the whole graph, while GNN encodes global features and Transformer network encodes local ones. Furthermore, previous works ignored encoding global interaction features of the heterogeneous graph with separate networks, thus leading to suboptimal performance. In this work, we present a novel framework called Relation-Aware GNN with transFormer (RAGFormer) which simultaneously embeds local and global features into a target node. The simple yet effective network applies a modified GAGA module where each transformer layer is followed by a cross-relation aggregation layer, to encode local embeddings and node interactions across different relations. Apart from the Transformer-based network, we further introduce a Relation-Aware GNN module to learn global embeddings, which is later merged into the local embeddings by an attention fusion module and a skip connection. Extensive experiments on two popular public datasets and an industrial dataset demonstrate that RAGFormer achieves the state-of-the-art performance. Substantial analysis experiments validate the effectiveness of each submodule of RAGFormer and its high efficiency in utilizing small-scale data and low hyper-parameter sensitivity.
Recently the retrieval-augmented generation (RAG) paradigm has raised much attention for its potential in incorporating external knowledge into large language models (LLMs) without further training. While widely explored in natural language applications, its utilization in code generation remains under-explored. In this paper, we introduce Active Retrieval in Knowledge Soup (ARKS), an advanced strategy for generalizing large language models for code. In contrast to relying on a single source, we construct a knowledge soup integrating web search, documentation, execution feedback, and evolved code snippets. We employ an active retrieval strategy that iteratively refines the query and updates the knowledge soup. To assess the performance of ARKS, we compile a new benchmark comprising realistic coding problems associated with frequently updated libraries and long-tail programming languages. Experimental results on ChatGPT and CodeLlama demonstrate a substantial improvement in the average execution accuracy of ARKS on LLMs. The analysis confirms the effectiveness of our proposed knowledge soup and active retrieval strategies, offering rich insights into the construction of effective retrieval-augmented code generation (RACG) pipelines. Our model, code, and data are available at https://arks-codegen.github.io.
Efficient attentions have greatly improved the computational efficiency of Transformers. However, most existing linear attention mechanisms suffer from an \emph{efficiency degradation} problem, leading to inefficiencies in causal language modeling and hindering their application in long-range language models. This problem is more pronounced under language modeling with unbounded contexts. In this paper, we propose \textbf{L}inear \textbf{A}ttention \textbf{V}ia \textbf{O}rthogonal memory~(\shortname) to address these limitations, achieving strong performance while maintaining linear complexity. \shortname employs orthogonal decomposition to compress a context into a fixed-size orthogonal memory while effectively minimizing redundancy within the context. Given that orthogonal memory compresses global information, we further dissect the context to amplify fine-grained local information. Additionally, we embed the relative position encoding into \shortname to improve the extrapolation ability. Experimental results show that \shortname greatly improves the efficiency of the causal language model with the best extrapolation performance and outperforms other efficient baselines. Further, we endeavor to employ \shortname for unbounded language modeling and successfully scale the context length to 128K.
In this paper, the recursive least squares (RLS) algorithm is considered in the sparse system identification setting. The cost function of RLS algorithm is regularized by a $p$-norm-like ($0 \leq p \leq 1$) constraint of the estimated system parameters. In order to minimize the regularized cost function, we transform it into a penalized maximum likelihood (ML) problem, which is solved by the expectation-maximization (EM) algorithm. With the introduction of a thresholding operator, the update equation of the tap-weight vector is derived. We also exploit the underlying sparsity to implement the proposed algorithm in a low computational complexity fashion. Numerical simulations demonstrate the superiority of the new algorithm over conventional sparse RLS algorithms, as well as regular RLS algorithm.
Autoregressive~(AR) generation almost dominates sequence generation for its efficacy. Recently, non-autoregressive~(NAR) generation gains increasing popularity for its efficiency and growing efficacy. However, its efficiency is still bottlenecked by quadratic complexity in sequence lengths, which is prohibitive for scaling to long sequence generation and few works have been done to mitigate this problem. In this paper, we propose a novel MLP variant, \textbf{A}ttentive \textbf{M}ulti-\textbf{L}ayer \textbf{P}erceptron~(AMLP), to produce a generation model with linear time and space complexity. Different from classic MLP with static and learnable projection matrices, AMLP leverages adaptive projections computed from inputs in an attentive mode. The sample-aware adaptive projections enable communications among tokens in a sequence, and model the measurement between the query and key space. Furthermore, we marry AMLP with popular NAR models, deriving a highly efficient NAR-AMLP architecture with linear time and space complexity. Empirical results show that such marriage architecture surpasses competitive efficient NAR models, by a significant margin on text-to-speech synthesis and machine translation. We also test AMLP's self- and cross-attention ability separately with extensive ablation experiments, and find them comparable or even superior to the other efficient models. The efficiency analysis further shows that AMLP extremely reduces the memory cost against vanilla non-autoregressive models for long sequences.
Large language models (LLMs) have already revolutionized code generation, after being pretrained on publicly available code data. However, while various methods have been proposed to augment LLMs with retrieved knowledge and enhance the quality of code generation, the performance of these retrieval-based methods is limited by the strength of the retrievers used. In addition, while LLMs show great emergent ability, they still struggle to produce the correct code in one turn. To address these challenges, we propose a novel two-step pipeline, called \autoknow, that leverages LLMs as both knowledge providers and self-reflective programmers. Unlike retrieval-based methods, \autoknow~obtains the knowledge from input prompts and generates intermediate code based on the generated knowledge. After that, \autoknow~asks LLM to act as an expert programmer to perform debugging for the generated code. This is achieved by receiving the error message from the interpreter, without requiring special test cases for correctness verification. We evaluate \autoknow~on three code generation datasets, including DS-1000 for data science code, HumanEval for software engineering code, and TransCoder for C++-to-Python translation. Our empirical experiments show that \autoknow~outperforms strong baselines by a significant margin on all datasets. We also conduct exhaustive analytical experiments to validate the effectiveness of the two stages of \autoknow, and find that both are superior to other prompting-based methods. Further scalability analysis demonstrates that \autoknow~can be adapted to other more advanced models, such as GPT-4, and bring consistent efficacy improvement.
Transformer has achieved remarkable success in language, image, and speech processing. Recently, various efficient attention architectures have been proposed to improve transformer's efficiency while largely preserving its efficacy, especially in modeling long sequences. A widely-used benchmark to test these efficient methods' capability on long-range modeling is Long Range Arena (LRA). However, LRA only focuses on the standard bidirectional (or noncausal) self attention, and completely ignores cross attentions and unidirectional (or causal) attentions, which are equally important to downstream applications. Although designing cross and causal variants of an attention method is straightforward for vanilla attention, it is often challenging for efficient attentions with subquadratic time and memory complexity. In this paper, we propose Comprehensive Attention Benchmark (CAB) under a fine-grained attention taxonomy with four distinguishable attention patterns, namely, noncausal self, causal self, noncausal cross, and causal cross attentions. CAB collects seven real-world tasks from different research areas to evaluate efficient attentions under the four attention patterns. Among these tasks, CAB validates efficient attentions in eight backbone networks to show their generalization across neural architectures. We conduct exhaustive experiments to benchmark the performances of nine widely-used efficient attention architectures designed with different philosophies on CAB. Extensive experimental results also shed light on the fundamental problems of efficient attentions, such as efficiency length against vanilla attention, performance consistency across attention patterns, the benefit of attention mechanisms, and interpolation/extrapolation on long-context language modeling.
An efficient 3D point cloud learning architecture, named PWCLO-Net, for LiDAR odometry is first proposed in this paper. In this architecture, the projection-aware representation of the 3D point cloud is proposed to organize the raw 3D point cloud into an ordered data form to achieve efficiency. The Pyramid, Warping, and Cost volume (PWC) structure for the LiDAR odometry task is built to estimate and refine the pose in a coarse-to-fine approach hierarchically and efficiently. A projection-aware attentive cost volume is built to directly associate two discrete point clouds and obtain embedding motion patterns. Then, a trainable embedding mask is proposed to weigh the local motion patterns to regress the overall pose and filter outlier points. The trainable pose warp-refinement module is iteratively used with embedding mask optimized hierarchically to make the pose estimation more robust for outliers. The entire architecture is holistically optimized end-to-end to achieve adaptive learning of cost volume and mask, and all operations involving point cloud sampling and grouping are accelerated by projection-aware 3D feature learning methods. The superior performance and effectiveness of our LiDAR odometry architecture are demonstrated on KITTI odometry dataset. Our method outperforms all recent learning-based methods and even the geometry-based approach, LOAM with mapping optimization, on most sequences of KITTI odometry dataset.