This paper describes the systems submitted by team6 for ChatEval, the DSTC 11 Track 4 competition. We present three different approaches to predicting turn-level qualities of chatbot responses based on large language models (LLMs). We report improvement over the baseline using dynamic few-shot examples from a vector store for the prompts for ChatGPT. We also analyze the performance of the other two approaches and report needed improvements for future work. We developed the three systems over just two weeks, showing the potential of LLMs for this task. An ablation study conducted after the challenge deadline shows that the new Llama 2 models are closing the performance gap between ChatGPT and open-source LLMs. However, we find that the Llama 2 models do not benefit from few-shot examples in the same way as ChatGPT.
相關內容
We consider the diffusion of two alternatives in social networks using a game-theoretic approach. Each individual plays a coordination game with its neighbors repeatedly and decides which to adopt. As products are used in conjunction with others and through repeated interactions, individuals are more interested in their long-term benefits and tend to show trust to others to maximize their long-term utility by choosing a suboptimal option with respect to instantaneous payoff. To capture such trust behavior, we deploy limited-trust equilibrium (LTE) in diffusion process. We analyze the convergence of emerging dynamics to equilibrium points using mean-field approximation and study the equilibrium state and the convergence rate of diffusion using absorption probability and expected absorption time of a reduced-size absorbing Markov chain. We also show that the diffusion model on LTE under the best-response strategy can be converted to the well-known linear threshold model. Simulation results show that when agents behave trustworthy, their long-term utility will increase significantly compared to the case when they are solely self-interested. Moreover, the Markov chain analysis provides a good estimate of convergence properties over random networks.
Bill James' Pythagorean formula has for decades done an excellent job estimating a team's winning percentage from very little data: if the average runs scored and allowed are denoted respectively by ${\rm RS}$ and ${\rm RA}$, there is some $\gamma$ such that the winning percentage is approximately ${\rm RS}^\gamma / ({\rm RS}^\gamma + {\rm RA}^\gamma)$. One important consequence is to determine the value of different players to the team, as it allows us to estimate how many more wins we would have given a fixed increase in run production. We summarize earlier work on the subject, and extend the earlier theoretical model of Miller (who estimated the run distributions as arising from independent Weibull distributions with the same shape parameter; this has been observed to describe the observed run data well). We now model runs scored and allowed as being drawn from independent Weibull distributions where the shape parameter is not necessarily the same, and then using the Method of Moments to solve a system of four equations in four unknowns. Doing so yields a predicted winning percentage that is often better than earlier models. This comes at a small cost as we no longer have a closed form expression but must evaluate a two-dimensional integral of two Weibull distributions and numerically estimate the solutions to the system of equations; as these are trivial to do with simple computational programs it is well worth adopting this framework and avoiding the issues of implementing the Method of Least Squares or the Method of Maximum Likelihood.
To maintain a reliable grid we need fast decision-making algorithms for complex problems like Dynamic Reconfiguration (DyR). DyR optimizes distribution grid switch settings in real-time to minimize grid losses and dispatches resources to supply loads with available generation. DyR is a mixed-integer problem and can be computationally intractable to solve for large grids and at fast timescales. We propose GraPhyR, a Physics-Informed Graph Neural Network (GNNs) framework tailored for DyR. We incorporate essential operational and connectivity constraints directly within the GNN framework and train it end-to-end. Our results show that GraPhyR is able to learn to optimize the DyR task.
While VideoQA Transformer models demonstrate competitive performance on standard benchmarks, the reasons behind their success are not fully understood. Do these models jointly capture and leverage the rich multimodal structures and dynamics from video and text? Or are they merely exploiting shortcuts to achieve high scores? Hence, we design $\textit{QUAG}$ (QUadrant AveraGe), a lightweight and non-parametric probe, to critically analyze multimodal representations. QUAG facilitates combined dataset-model study by systematic ablation of model's coupled multimodal understanding during inference. Surprisingly, it demonstrates that the models manage to maintain high performance even under multimodal impairment. We extend QUAG to design "QUAG-attention", a simplistic and less-expressive replacement of self-attention. We find that the models with QUAG-attention achieve similar performance with significantly less mulops without any finetuning. These findings indicate that the current VideoQA benchmarks and metrics do not penalize models that find shortcuts and discount joint multimodal understanding. Motivated by this, we propose the $\textit{CLAVI}$ (Counterfactual in LAnguage and VIdeo), a diagnostic dataset for coupled multimodal understanding in VideoQA. CLAVI consists of temporal questions and videos that are augmented to curate balanced counterfactuals in language and video domains. We evaluate models on CLAVI and find that all models achieve high performance on multimodal shortcut instances, but most of them have poor performance on the counterfactual instances that necessitate joint multimodal understanding. Overall, with the multimodal representation analysis using QUAG and diagnostic analysis using CLAVI, we show that many VideoQA models are incapable of learning multimodal representations and that their success on standard datasets is an illusion of joint multimodal understanding.
This paper performs the first study to understand the prevalence, challenges, and effectiveness of using Static Application Security Testing (SAST) tools on Open-Source Embedded Software (EMBOSS) repositories. We collect a corpus of 258 of the most popular EMBOSS projects, representing 13 distinct categories such as real-time operating systems, network stacks, and applications. To understand the current use of SAST tools on EMBOSS, we measured this corpus and surveyed developers. To understand the challenges and effectiveness of using SAST tools on EMBOSS projects, we applied these tools to the projects in our corpus. We report that almost none of these projects (just 3%) use SAST tools beyond those baked into the compiler, and developers give rationales such as ineffectiveness and false positives. In applying SAST tools ourselves, we show that minimal engineering effort and project expertise are needed to apply many tools to a given EMBOSS project. GitHub's CodeQL was the most effective SAST tool -- using its built-in security checks we found a total of 540 defects (with a false positive rate of 23%) across the 258 projects, with 399 (74%) likely security vulnerabilities, including in projects maintained by Microsoft, Amazon, and the Apache Foundation. EMBOSS engineers have confirmed 273 (51%) of these defects, mainly by accepting our pull requests. Two CVEs were issued. In summary, we urge EMBOSS engineers to adopt the current generation of SAST tools, which offer low false positive rates and are effective at finding security-relevant defects.
This work investigates the potential of Federated Learning (FL) for official statistics and shows how well the performance of FL models can keep up with centralized learning methods.F L is particularly interesting for official statistics because its utilization can safeguard the privacy of data holders, thus facilitating access to a broader range of data. By simulating three different use cases, important insights on the applicability of the technology are gained. The use cases are based on a medical insurance data set, a fine dust pollution data set and a mobile radio coverage data set - all of which are from domains close to official statistics. We provide a detailed analysis of the results, including a comparison of centralized and FL algorithm performances for each simulation. In all three use cases, we were able to train models via FL which reach a performance very close to the centralized model benchmarks. Our key observations and their implications for transferring the simulations into practice are summarized. We arrive at the conclusion that FL has the potential to emerge as a pivotal technology in future use cases of official statistics.
Recently, significant progress has been made in understanding the generalization of neural networks (NNs) trained by gradient descent (GD) using the algorithmic stability approach. However, most of the existing research has focused on one-hidden-layer NNs and has not addressed the impact of different network scaling parameters. In this paper, we greatly extend the previous work \cite{lei2022stability,richards2021stability} by conducting a comprehensive stability and generalization analysis of GD for multi-layer NNs. For two-layer NNs, our results are established under general network scaling parameters, relaxing previous conditions. In the case of three-layer NNs, our technical contribution lies in demonstrating its nearly co-coercive property by utilizing a novel induction strategy that thoroughly explores the effects of over-parameterization. As a direct application of our general findings, we derive the excess risk rate of $O(1/\sqrt{n})$ for GD algorithms in both two-layer and three-layer NNs. This sheds light on sufficient or necessary conditions for under-parameterized and over-parameterized NNs trained by GD to attain the desired risk rate of $O(1/\sqrt{n})$. Moreover, we demonstrate that as the scaling parameter increases or the network complexity decreases, less over-parameterization is required for GD to achieve the desired error rates. Additionally, under a low-noise condition, we obtain a fast risk rate of $O(1/n)$ for GD in both two-layer and three-layer NNs.
Pre-trained Language Models (PLMs) have achieved great success in various Natural Language Processing (NLP) tasks under the pre-training and fine-tuning paradigm. With large quantities of parameters, PLMs are computation-intensive and resource-hungry. Hence, model pruning has been introduced to compress large-scale PLMs. However, most prior approaches only consider task-specific knowledge towards downstream tasks, but ignore the essential task-agnostic knowledge during pruning, which may cause catastrophic forgetting problem and lead to poor generalization ability. To maintain both task-agnostic and task-specific knowledge in our pruned model, we propose ContrAstive Pruning (CAP) under the paradigm of pre-training and fine-tuning. It is designed as a general framework, compatible with both structured and unstructured pruning. Unified in contrastive learning, CAP enables the pruned model to learn from the pre-trained model for task-agnostic knowledge, and fine-tuned model for task-specific knowledge. Besides, to better retain the performance of the pruned model, the snapshots (i.e., the intermediate models at each pruning iteration) also serve as effective supervisions for pruning. Our extensive experiments show that adopting CAP consistently yields significant improvements, especially in extremely high sparsity scenarios. With only 3% model parameters reserved (i.e., 97% sparsity), CAP successfully achieves 99.2% and 96.3% of the original BERT performance in QQP and MNLI tasks. In addition, our probing experiments demonstrate that the model pruned by CAP tends to achieve better generalization ability.
Properly handling missing data is a fundamental challenge in recommendation. Most present works perform negative sampling from unobserved data to supply the training of recommender models with negative signals. Nevertheless, existing negative sampling strategies, either static or adaptive ones, are insufficient to yield high-quality negative samples --- both informative to model training and reflective of user real needs. In this work, we hypothesize that item knowledge graph (KG), which provides rich relations among items and KG entities, could be useful to infer informative and factual negative samples. Towards this end, we develop a new negative sampling model, Knowledge Graph Policy Network (KGPolicy), which works as a reinforcement learning agent to explore high-quality negatives. Specifically, by conducting our designed exploration operations, it navigates from the target positive interaction, adaptively receives knowledge-aware negative signals, and ultimately yields a potential negative item to train the recommender. We tested on a matrix factorization (MF) model equipped with KGPolicy, and it achieves significant improvements over both state-of-the-art sampling methods like DNS and IRGAN, and KG-enhanced recommender models like KGAT. Further analyses from different angles provide insights of knowledge-aware sampling. We release the codes and datasets at //github.com/xiangwang1223/kgpolicy.
Deep Convolutional Neural Networks (CNNs) are a special type of Neural Networks, which have shown state-of-the-art results on various competitive benchmarks. The powerful learning ability of deep CNN is largely achieved with the use of multiple non-linear feature extraction stages that can automatically learn hierarchical representation from the data. Availability of a large amount of data and improvements in the hardware processing units have accelerated the research in CNNs and recently very interesting deep CNN architectures are reported. The recent race in deep CNN architectures for achieving high performance on the challenging benchmarks has shown that the innovative architectural ideas, as well as parameter optimization, can improve the CNN performance on various vision-related tasks. In this regard, different ideas in the CNN design have been explored such as use of different activation and loss functions, parameter optimization, regularization, and restructuring of processing units. However, the major improvement in representational capacity is achieved by the restructuring of the processing units. Especially, the idea of using a block as a structural unit instead of a layer is gaining substantial appreciation. This survey thus focuses on the intrinsic taxonomy present in the recently reported CNN architectures and consequently, classifies the recent innovations in CNN architectures into seven different categories. These seven categories are based on spatial exploitation, depth, multi-path, width, feature map exploitation, channel boosting and attention. Additionally, it covers the elementary understanding of the CNN components and sheds light on the current challenges and applications of CNNs.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191