Search

This paper revisits recent code similarity evaluation metrics, particularly focusing on the application of Abstract Syntax Tree (AST) editing distance in diverse programming languages. In particular, we explore the usefulness of these metrics and compare them to traditional sequence similarity metrics. Our experiments showcase the effectiveness of AST editing distance in capturing intricate code structures, revealing a high correlation with established metrics. Furthermore, we explore the strengths and weaknesses of AST editing distance and prompt-based GPT similarity scores in comparison to BLEU score, execution match, and Jaccard Similarity. We propose, optimize, and publish an adaptable metric that demonstrates effectiveness across all tested languages, representing an enhanced version of Tree Similarity of Edit Distance (TSED)., Comment: ACL 2024 Main

Code review, which aims at ensuring the overall quality and reliability of software, is a cornerstone of software development. Unfortunately, while crucial, Code review is a labor-intensive process that the research community is looking to automate. Existing automated methods rely on single input-output generative models and thus generally struggle to emulate the collaborative nature of code review. This work introduces CodeAgent, a novel multi-agent Large Language Model (LLM) system for code review automation. CodeAgent incorporates a supervisory agent, QA-Checker, to ensure that all the agents' contributions address the initial review question. We evaluated CodeAgent on critical code review tasks: (1) detect inconsistencies between code changes and commit messages, (2) identify vulnerability introductions, (3) validate code style adherence, and (4) suggest code revisions. The results demonstrate CodeAgent's effectiveness, contributing to a new state-of-the-art in code review automation. Our data and code are publicly available (\url{https://github.com/Code4Agent/codeagent}).

Multi-messenger searches for BNS and NSBH mergers are currently one of the most exciting areas of astronomy. The search for joint electromagnetic and neutrino counterparts to GWs has resumed with O4. To support this effort, public semi-automated data products are sent in near real-time and include localization and source properties to guide complementary observations. In preparation for O4, we have conducted a study using a simulated population of compact binaries and a MDC in the form of a real-time replay to optimize and profile the software infrastructure and scientific deliverables. End-to-end performance was tested, including data ingestion, running online search pipelines, performing annotations, and issuing alerts to the astrophysics community. We present an overview of the low-latency infrastructure and the performance of the data products that are now being released during O4 based on the MDC. We report the expected median latency for the preliminary alert of full bandwidth searches (29.5s) and show consistency and accuracy of released data products using the MDC. For the first time, we report the expected median latency for triggers from early warning searches (-3.1s), which are new in O4 and target neutron star mergers during inspiral phase. This paper provides a performance overview for LVK low-latency alert infrastructure and data products using the MDC and serves as a useful reference for the interpretation of O4 detections.

We leverage different context windows when predicting the emotion of different utterances. New modules are included to realize variable-length context: 1) two speaker-aware units, which explicitly model inner- and inter-speaker dependencies to form distilled conversational context, and 2) a top-k normalization layer, which determines the most proper context windows from the conversational context to predict emotion. Experiments and ablation studies show that our approach outperforms several strong baselines on three public datasets.

International Classification of Diseases (ICD) is a set of classification codes for medical records. Automated ICD coding, which assigns unique International Classification of Diseases codes with each medical record, is widely used recently for its efficiency and error-prone avoidance. However, there are challenges that remain such as heterogeneity, label unbalance, and complex relationships between ICD codes. In this work, we proposed a novel Bidirectional Hierarchy Framework(HieNet) to address the challenges. Specifically, a personalized PageRank routine is developed to capture the co-relation of codes, a bidirectional hierarchy passage encoder to capture the codes' hierarchical representations, and a progressive predicting method is then proposed to narrow down the semantic searching space of prediction. We validate our method on two widely used datasets. Experimental results on two authoritative public datasets demonstrate that our proposed method boosts state-of-the-art performance by a large margin.

Medical dialogue information extraction is becoming an increasingly significant problem in modern medical care. It is difficult to extract key information from electronic medical records (EMRs) due to their large numbers. Previously, researchers proposed attention-based models for retrieving features from EMRs, but their limitations were reflected in their inability to recognize different categories in medical dialogues. In this paper, we propose a novel model, Expert System and Attention for Labelling (ESAL). We use mixture of experts and pre-trained BERT to retrieve the semantics of different categories, enabling the model to fuse the differences between them. In our experiment, ESAL was applied to a public dataset and the experimental results indicated that ESAL significantly improved the performance of Medical Information Classification.

Every day, weather forecasting centres around the world make use of noisy, incomplete observations of the atmosphere to update their weather forecasts. This process is known as data assimilation, data fusion or state estimation and is best expressed as Bayesian inference: given a set of observations, some prior beliefs and a model of the target system, what is the probability distribution of some set of unobserved quantities or latent variables at some time, possibly in the future? While data assimilation has developed rapidly in some areas, relatively little progress has been made in performing data assimilation with agent-based models. This has hampered the use of agent-based models to make quantitative claims about real-world systems. Here we present an algorithm that uses Markov-Chain-Monte-Carlo methods to generate samples of the parameters and trajectories of an agent-based model over a window of time given a set of possibly noisy, aggregated and incomplete observations of the system. This can be used as-is, or as part of a data assimilation cycle or sequential-MCMC algorithm. Our algorithm is applicable to time-stepping, agent-based models whose agents have a finite set of states and a finite number of ways of acting on the world. As presented the algorithm is only practical for agents with a few bytes of internal state although we discuss ways of removing this restriction. We demonstrate the algorithm by performing data assimilation with an agent-based, spatial predator-prey model., Comment: 22 pages, 4 figures

Many countries are currently gearing up to use smart-phone apps to perform contact tracing as part of the effort to manage the COVID-19 pandemic and prevent resurgences of the disease after the initial outbreak. With the announcement of the Apple/Google partnership to introduce contact-tracing functionality to iOS and Android, it seems likely that this will be adopted in many countries. An important part of the functionality of the app will be to decide whether a person should be advised to self-isolate, be tested or end isolation. However, the privacy preserving nature of the Apple/Google contact tracing algorithm means that centralised curation of these decisions is not possible so each phone must use its own "risk model" to inform decisions. Ideally, the risk model should use Bayesian inference to decide the best course of action given the test results of the user and those of other users. Here we present a decentralised algorithm that estimates the Bayesian posterior probability of viral transmission events and evaluates when a user should be notified, tested or released from isolation while preserving user privacy. The algorithm also allows the disease models on the phones to learn from everyone's contact-tracing data and will allow Epidemiologists to better understand the dynamics of the disease. The algorithm is a message passing algorithm, based on belief propagation, so each smart-phone can be used to execute a small part of the algorithm without releasing any sensitive information. In this way, the network of all participating smart-phones forms a distributed computation device that performs Bayesian inference, informs each user when they should start/end isolation or be tested and learns about the disease from user's data.

In this paper we consider the problem of finding the most probable set of events that could have led to a set of partial, noisy observations of some dynamical system. In particular, we consider the case of a dynamical system that is a (possibly stochastic) time-stepping agent-based model with a discrete state space, the (possibly noisy) observations are the number of agents that have some given property and the events we're interested in are the decisions made by the agents (their ``expressed behaviours'') as the model evolves. We show that this problem can be reduced to an integer linear programming problem which can subsequently be solved numerically using a standard branch-and-cut algorithm. We describe two implementations, an ``offline'' algorithm that finds the maximum-a-posteriori expressed behaviours given a set of observations over a finite time window, and an ``online'' algorithm that incrementally builds a feasible set of behaviours from a stream of observations that may have no natural beginning or end. We demonstrate both algorithms on a spatial predator-prey model on a 32x32 grid with an initial population of 100 agents.

Agent-based models are a powerful tool for studying the behaviour of complex systems that can be described in terms of multiple, interacting ``agents''. However, because of their inherently discrete and often highly non-linear nature, it is very difficult to reason about the relationship between the state of the model, on the one hand, and our observations of the real world on the other. In this paper we consider agents that have a discrete set of states that, at any instant, act with a probability that may depend on the environment or the state of other agents. Given this, we show how the mathematical apparatus of quantum field theory can be used to reason probabilistically about the state and dynamics the model, and describe an algorithm to update our belief in the state of the model in the light of new, real-world observations. Using a simple predator-prey model on a 2-dimensional spatial grid as an example, we demonstrate the assimilation of incomplete, noisy observations and show that this leads to an increase in the mutual information between the actual state of the observed system and the posterior distribution given the observations, when compared to a null model.

Probabilistic programming languages and other machine learning applications often require samples to be generated from a categorical distribution where the probability of each one of $n$ categories is specified as a parameter. If the parameters are hyper-parameters then they need to be modified, however, current implementations of categorical distributions take $\mathcal{O}(n)$ time to modify a parameter. If $n$ is large and the parameters are being frequently modified, this can become prohibitive. Here we present the insight that a Huffman tree is an efficient data structure for representing categorical distributions and present algorithms to generate samples as well as add, delete and modify categories in $\mathcal{O}(\log(n))$ time. We demonstrate that the time to sample from the distribution remains, in practice, within a few percent of the theoretical optimal value. The same algorithm may also be useful in the context of adaptive Huffman coding where computational efficiency is important.