In this work, we address the problem of directing the text generation of a language model (LM) towards a desired behavior, aligning the generated text with the preferences of the human operator. We propose using another, instruction-tuned language model as a critic reward model in a zero-shot way thanks to the prompt of a Yes-No question that represents the user preferences, without requiring further labeled data. This zero-shot reward model provides the learning signal to further fine-tune the base LM using Reinforcement Learning from AI Feedback (RLAIF); yet our approach is also compatible in other contexts such as quality-diversity search. Extensive evidence of the capabilities of the proposed ZYN framework is provided through experiments in different domains related to text generation, including detoxification; optimizing sentiment of movie reviews, or any other attribute; steering the opinion about a particular topic the model may have; and personalizing prompt generators for text-to-image tasks. Code available at \url{//github.com/vicgalle/zero-shot-reward-models/}.
相關內容
ACM/IEEE第23屆模型驅動工程語言和系統國際會議，是模型驅動軟件和系統工程的首要會議系列，由ACM-SIGSOFT和IEEE-TCSE支持組織。自1998年以來，模型涵蓋了建模的各個方面，從語言和方法到工具和應用程序。模特的參加者來自不同的背景，包括研究人員、學者、工程師和工業專業人士。MODELS 2019是一個論壇，參與者可以圍繞建模和模型驅動的軟件和系統交流前沿研究成果和創新實踐經驗。今年的版本將為建模社區提供進一步推進建模基礎的機會，并在網絡物理系統、嵌入式系統、社會技術系統、云計算、大數據、機器學習、安全、開源等新興領域提出建模的創新應用以及可持續性。
官網鏈接： ·
MoDELS
·
語言模型化
·
大語言模型
·
可約的
·
Despite the impressive capabilities of large language models (LLMs) across diverse applications, they still suffer from trustworthiness issues, such as hallucinations and misalignments. Retrieval-augmented language models (RAG) have been proposed to enhance the credibility of generations by grounding external knowledge, but the theoretical understandings of their generation risks remains unexplored. In this paper, we answer: 1) whether RAG can indeed lead to low generation risks, 2) how to provide provable guarantees on the generation risks of RAG and vanilla LLMs, and 3) what sufficient conditions enable RAG models to reduce generation risks. We propose C-RAG, the first framework to certify generation risks for RAG models. Specifically, we provide conformal risk analysis for RAG models and certify an upper confidence bound of generation risks, which we refer to as conformal generation risk. We also provide theoretical guarantees on conformal generation risks for general bounded risk functions under test distribution shifts. We prove that RAG achieves a lower conformal generation risk than that of a single LLM when the quality of the retrieval model and transformer is non-trivial. Our intensive empirical results demonstrate the soundness and tightness of our conformal generation risk guarantees across four widely-used NLP datasets on four state-of-the-art retrieval models.
In this work, we present a comprehensive exploration of finetuning Malaysian language models, specifically Llama2 and Mistral, on embedding tasks involving negative and positive pairs. We release two distinct models tailored for Semantic Similarity and Retrieval-Augmented Generation (RAG). For Semantic Similarity, our 600 million parameter Llama2 model outperforms OpenAI text-embedding-ada-002 across all recall@k metrics for b.cari.com.my, c.cari.com.my, Malay news, and Malaysian Twitter test sets. In the realm of RAG models, our approach proves competitive with OpenAI text-embedding-ada-002 in the Malaysian context. Notably, our 2 billion parameter Llama2 model achieves superior Recall@5, Recall@10 for the "Melayu" keyword research papers dataset and excels in Recall@3, Recall@5, and Recall@10 for the lom.agc.gov.my dataset. These findings underscore the effectiveness of our finetuning strategy and highlight the performance gains in both Semantic Similarity and RAG tasks. All models released at //huggingface.co/collections/mesolitica/malaysian-embedding-6523612bfe5881ad35f81b99
In this work we present FreDSNet, a deep learning solution which obtains semantic 3D understanding of indoor environments from single panoramas. Omnidirectional images reveal task-specific advantages when addressing scene understanding problems due to the 360-degree contextual information about the entire environment they provide. However, the inherent characteristics of the omnidirectional images add additional problems to obtain an accurate detection and segmentation of objects or a good depth estimation. To overcome these problems, we exploit convolutions in the frequential domain obtaining a wider receptive field in each convolutional layer. These convolutions allow to leverage the whole context information from omnidirectional images. FreDSNet is the first network that jointly provides monocular depth estimation and semantic segmentation from a single panoramic image exploiting fast Fourier convolutions. Our experiments show that FreDSNet has similar performance as specific state of the art methods for semantic segmentation and depth estimation. FreDSNet code is publicly available in //github.com/Sbrunoberenguel/FreDSNet
In this paper, we study the stochastic collocation (SC) methods for uncertainty quantification (UQ) in hyperbolic systems of nonlinear partial differential equations (PDEs). In these methods, the underlying PDEs are numerically solved at a set of collocation points in random space. A standard SC approach is based on a generalized polynomial chaos (gPC) expansion, which relies on choosing the collocation points based on the prescribed probability distribution and approximating the computed solution by a linear combination of orthogonal polynomials in the random variable. We demonstrate that this approach struggles to accurately capture discontinuous solutions, often leading to oscillations (Gibbs phenomenon) that deviate significantly from the physical solutions. We explore alternative SC methods, in which one can choose an arbitrary set of collocation points and employ shape-preserving splines to interpolate the solution in a random space. Our study demonstrates the effectiveness of spline-based collocation in accurately capturing and assessing uncertainties while suppressing oscillations. We illustrate the superiority of the spline-based collocation on two numerical examples, including the inviscid Burgers and shallow water equations.
In this research, we present SLYKLatent, a novel approach for enhancing gaze estimation by addressing appearance instability challenges in datasets due to aleatoric uncertainties, covariant shifts, and test domain generalization. SLYKLatent utilizes Self-Supervised Learning for initial training with facial expression datasets, followed by refinement with a patch-based tri-branch network and an inverse explained variance-weighted training loss function. Our evaluation on benchmark datasets achieves an 8.7% improvement on Gaze360, rivals top MPIIFaceGaze results, and leads on a subset of ETH-XGaze by 13%, surpassing existing methods by significant margins. Adaptability tests on RAF-DB and Affectnet show 86.4% and 60.9% accuracies, respectively. Ablation studies confirm the effectiveness of SLYKLatent's novel components. This approach has strong potential in human-robot interaction.
In recent years, there has been a lot of research work activity focused on carrying out asymptotic and non-asymptotic convergence analyses for two-timescale actor critic algorithms where the actor updates are performed on a timescale that is slower than that of the critic. In a recent work, the critic-actor algorithm has been presented for the infinite horizon discounted cost setting in the look-up table case where the timescales of the actor and the critic are reversed and asymptotic convergence analysis has been presented. In our work, we present the first critic-actor algorithm with function approximation and in the long-run average reward setting and present the first finite-time (non-asymptotic) analysis of such a scheme. We obtain optimal learning rates and prove that our algorithm achieves a sample complexity of $\mathcal{\tilde{O}}(\epsilon^{-2.08})$ for the mean squared error of the critic to be upper bounded by $\epsilon$ which is better than the one obtained for actor-critic in a similar setting. We also show the results of numerical experiments on three benchmark settings and observe that the critic-actor algorithm competes well with the actor-critic algorithm.
Safety in the face of uncertainty is a key challenge in robotics. In this work, we propose a real-time capable framework to generate safe and task-efficient robot trajectories for stochastic control problems. For that, we first formulate the problem as a chance-constrained optimisation problem, in which the probability of the controlled system to violate a safety constraint is constrained to be below a user-defined threshold. To solve the chance-constrained optimisation problem, we propose a Monte--Carlo approximation relying on samples of the uncertainty to estimate the probability of violating a safety constraint given a controller. We use this approximation in the motion planner VP-STO to solve the sampled-based problem. Consequently, we refer to our adapted approach as CC-VPSTO, which stands for Chance-Constrained VP-STO. We address the crucial issue concerning the Monte--Carlo approximation: given a predetermined number of uncertainty samples, we propose several ways to define the sample-based problem such that it is a reliable over-approximation of the original problem, i.e. any solution to the sample-based problem adheres to the original chance-constrained problem with high confidence. The strengths of our approach lie in i) its generality, as it does not require any specific assumptions on the underlying uncertainty distribution, the dynamics of the system, the cost function, and for some of the proposed sample-based approximations, on the form of inequality constraints; and ii) its applicability to MPC-settings. We demonstrate the validity and efficiency of our approach on both simulation and real-world robot experiments. For additional material, please visit //sites.google.com/oxfordrobotics.institute/cc-vpsto.
Recent research has made significant progress in designing fusion modules for audio-visual speech separation. However, they predominantly focus on multi-modal fusion at a single temporal scale of auditory and visual features without employing selective attention mechanisms, which is in sharp contrast with the brain. To address this issue, We propose a novel model called Intra- and Inter-Attention Network (IIANet), which leverages the attention mechanism for efficient audio-visual feature fusion. IIANet consists of two types of attention blocks: intra-attention (IntraA) and inter-attention (InterA) blocks, where the InterA blocks are distributed at the top, middle and bottom of IIANet. Heavily inspired by the way how human brain selectively focuses on relevant content at various temporal scales, these blocks maintain the ability to learn modality-specific features and enable the extraction of different semantics from audio-visual features. Comprehensive experiments on three standard audio-visual separation benchmarks (LRS2, LRS3, and VoxCeleb2) demonstrate the effectiveness of IIANet, outperforming previous state-of-the-art methods while maintaining comparable inference time. In particular, the fast version of IIANet (IIANet-fast) has only 7% of CTCNet's MACs and is 40% faster than CTCNet on CPUs while achieving better separation quality, showing the great potential of attention mechanism for efficient and effective multimodal fusion.
Pretrained language models (PLMs) have shown remarkable generalization toward multiple tasks and languages. Nonetheless, the generalization of PLMs towards unseen languages is poor, resulting in significantly worse language performance, or even generating nonsensical responses that are comparable to a random baseline. This limitation has been a longstanding problem of PLMs raising the problem of diversity and equal access to language modeling technology. In this work, we solve this limitation by introducing LinguAlchemy, a regularization technique that incorporates various aspects of languages covering typological, geographical, and phylogenetic constraining the resulting representation of PLMs to better characterize the corresponding linguistics constraints. LinguAlchemy significantly improves the accuracy performance of mBERT and XLM-R on unseen languages by ~18% and ~2%, respectively compared to fully finetuned models and displaying a high degree of unseen language generalization. We further introduce AlchemyScale and AlchemyTune, extension of LinguAlchemy which adjusts the linguistic regularization weights automatically, alleviating the need for hyperparameter search. LinguAlchemy enables better cross-lingual generalization to unseen languages which is vital for better inclusivity and accessibility of PLMs.
We study the problem of learning to reason in large scale knowledge graphs (KGs). More specifically, we describe a novel reinforcement learning framework for learning multi-hop relational paths: we use a policy-based agent with continuous states based on knowledge graph embeddings, which reasons in a KG vector space by sampling the most promising relation to extend its path. In contrast to prior work, our approach includes a reward function that takes the accuracy, diversity, and efficiency into consideration. Experimentally, we show that our proposed method outperforms a path-ranking based algorithm and knowledge graph embedding methods on Freebase and Never-Ending Language Learning datasets.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191