arXiv subjects

Tao Yu

Publications and source records attributed to Tao Yu.

At least 253 records · Page 14Linked to original sources

arXiv2022

Coder Reviewer Reranking for Code Generation

Sampling diverse programs from a code language model and reranking with model likelihood is a popular method for code generation but it is prone to preferring degenerate solutions. Inspired by collaborative programming, we propose Coder-Reviewer reranking. We augment Coder language models from past work, which generate programs given language instructions, with Reviewer models, which evaluate the likelihood of the instruction given the generated programs. We perform an extensive study across six datasets with eight models from three model families. Experimental results show that Coder-Reviewer reranking leads to consistent and significant improvement (up to 17% absolute accuracy gain) over reranking with the Coder model only. When combined with executability filtering, Coder-Reviewer reranking can often outperform the minimum Bayes risk method. Coder-Reviewer reranking is easy to implement by prompting, can generalize to different programming languages, and works well with off-the-shelf hyperparameters.

cs.LG↗

arXiv2022

Towards a Decentralized Metaverse: Synchronized Orchestration of Digital Twins and Sub-Metaverses

Accommodating digital twins (DTs) in the metaverse is essential to achieving digital reality. This need for integrating DTs into the metaverse while operating them at the network edge has increased the demand for a decentralized edge-enabled metaverse. Hence, to consolidate the fusion between real and digital entities, it is necessary to harmonize the interoperability between DTs and the metaverse at the edge. In this paper, a novel decentralized metaverse framework that incorporates DT operations at the wireless edge is presented. In particular, a system of autonomous physical twins (PTs) operating in a massively-sensed zone is replicated as cyber twins (CTs) at the mobile edge computing (MEC) servers. To render the CTs' digital environment, this zone is partitioned and teleported as distributed sub-metaverses to the MEC servers. To guarantee seamless synchronization of the sub-metaverses and their associated CTs with the dynamics of the real world and PTs, respectively, this joint synchronization problem is posed as an optimization problem whose goal is to minimize the average sub-synchronization time between the real and digital worlds, while meeting the DT synchronization intensity requirements. To solve this problem, a novel iterative algorithm for joint sub-metaverse and DT association at the MEC servers is proposed. This algorithm exploits the rigorous framework of optimal transport theory so as to efficiently distribute the sub-metaverses and DTs, while considering the computing and communication resource allocations. Simulation results show that the proposed solution can orchestrate the interplay between DTs and sub-metaverses to achieve a 25.75 % reduction in the sub-synchronization time in comparison to the signal-to-noise ratio-based association scheme.

cs.IT↗

arXiv2022

DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation

We introduce DS-1000, a code generation benchmark with a thousand data science problems spanning seven Python libraries, such as NumPy and Pandas. Compared to prior works, DS-1000 incorporates three core features. First, our problems reflect diverse, realistic, and practical use cases since we collected them from StackOverflow. Second, our automatic evaluation is highly specific (reliable) -- across all Codex-002-predicted solutions that our evaluation accept, only 1.8% of them are incorrect; we achieve this with multi-criteria metrics, checking both functional correctness by running test cases and surface-form constraints by restricting API usages or keywords. Finally, we proactively defend against memorization by slightly modifying our problems to be different from the original StackOverflow source; consequently, models cannot answer them correctly by memorizing the solutions from pre-training. The current best public system (Codex-002) achieves 43.3% accuracy, leaving ample room for improvement. We release our benchmark at https://ds1000-code-gen.github.io.

cs.SE↗

arXiv2022

Speech Enhancement with Fullband-Subband Cross-Attention Network

FullSubNet has shown its promising performance on speech enhancement by utilizing both fullband and subband information. However, the relationship between fullband and subband in FullSubNet is achieved by simply concatenating the output of fullband model and subband units. It only supplements the subband units with a small quantity of global information and has not considered the interaction between fullband and subband. This paper proposes a fullband-subband cross-attention (FSCA) module to interactively fuse the global and local information and applies it to FullSubNet. This new framework is called as FS-CANet. Moreover, different from FullSubNet, the proposed FS-CANet optimize the fullband extractor by temporal convolutional network (TCN) blocks to further reduce the model size. Experimental results on DNS Challenge - Interspeech 2021 dataset show that the proposed FS-CANet outperforms other state-of-the-art speech enhancement approaches, and demonstrate the effectiveness of fullband-subband cross-attention.

cs.SD↗

arXiv2022

Experiment of Multi-UAV Full-Duplex System Equipped with Directional Antennas

One of the key enablers for the realization of a variety of unmanned aerial vehicle (UAV)-based systems is the high-performance communication system linking many UAVs and ground station. We have proposed a spectrum-efficient full-duplex directional-antennas-equipped multi-UAV communication system with low hardware complexity to address the issues of low spectrum efficiency caused by co-channel interference in areal channels. In this paper, by using the prototype system including UAVs and ground station, field experiments are carried out to confirm the feasibility and effectiveness of the proposed system's key feature, i.e., co-channel interference cancellation among UAVs by directional antennas and UAV relative position control, instead of energy-consuming dedicated self-interference cancellers on UAVs in traditional full-duplex systems. Both uplink and downlink performance are tested. Specially, in downlink experiment, channel power of interference between a pair of two UAVs is measured when UAVs are in different positional relationships. The experiment results agree well with the designs and confirm that the proposed system can greatly improve the system performance.

cs.IT↗

arXiv2022

In-Context Learning for Few-Shot Dialogue State Tracking

Collecting and annotating task-oriented dialogues is time-consuming and costly; thus, zero and few shot learning could greatly benefit dialogue state tracking (DST). In this work, we propose an in-context learning (ICL) framework for zero-shot and few-shot learning DST, where a large pre-trained language model (LM) takes a test instance and a few exemplars as input, and directly decodes the dialogue state without any parameter updates. To better leverage a tabular domain description in the LM prompt, we reformulate DST into a text-to-SQL problem. We also propose a novel approach to retrieve annotated dialogues as exemplars. Empirical results on MultiWOZ show that our method IC-DST substantially outperforms previous fine-tuned state-of-the-art models in few-shot settings. In addition, we test IC-DST in zero-shot settings, in which the model only takes a fixed task instruction as input, finding that it outperforms previous zero-shot methods by a large margin.

cs.CL↗

arXiv2022

ZeroGen: Efficient Zero-shot Learning via Dataset Generation

There is a growing interest in dataset generation recently due to the superior generative capacity of large pre-trained language models (PLMs). In this paper, we study a flexible and efficient zero-short learning method, \textsc{ZeroGen}. Given a zero-shot task, we first generate a dataset from scratch using PLMs in an unsupervised manner. Then, we train a tiny task model (e.g., LSTM) under the supervision of the synthesized dataset. This approach allows highly efficient inference as the final task model only has orders of magnitude fewer parameters comparing to PLMs (e.g., GPT2-XL). Apart from being annotation-free and efficient, we argue that \textsc{ZeroGen} can also provide useful insights from the perspective of data-free model-agnostic knowledge distillation, and unreferenced text generation evaluation. Experiments and analysis on different NLP tasks, namely, text classification, question answering, and natural language inference, show the effectiveness of \textsc{ZeroGen}.

cs.CL↗

arXiv2022

ProGen: Progressive Zero-shot Dataset Generation via In-context Feedback

Recently, dataset-generation-based zero-shot learning has shown promising results by training a task-specific model with a dataset synthesized from large pre-trained language models (PLMs). The final task-specific model often achieves compatible or even better performance than PLMs under the zero-shot setting, with orders of magnitude fewer parameters. However, synthetic datasets have their drawbacks. They have long been suffering from low-quality issues (e.g., low informativeness and redundancy). This explains why the massive synthetic data does not lead to better performance -- a scenario we would expect in the human-labeled data. To improve the quality of dataset synthesis, we propose a progressive zero-shot dataset generation framework, ProGen, which leverages the feedback from the task-specific model to guide the generation of new training data via in-context examples. Extensive experiments on five text classification datasets demonstrate the effectiveness of the proposed approach. We also show ProGen achieves on-par or superior performance with only 1\% synthetic dataset size compared to baseline methods without in-context feedback.

cs.CL↗

arXiv2022

Augmenting Multi-Turn Text-to-SQL Datasets with Self-Play

The task of context-dependent text-to-SQL aims to convert multi-turn user utterances to formal SQL queries. This is a challenging task due to both the scarcity of training data from which to learn complex contextual dependencies and to generalize to unseen databases. In this paper we explore augmenting the training datasets using self-play, which leverages contextual information to synthesize new interactions to adapt the model to new databases. We first design a SQL-to-text model conditioned on a sampled goal query, which represents a user's intent, that then converses with a text-to-SQL semantic parser to generate new interactions. We then filter the synthesized interactions and retrain the models with the augmented data. We find that self-play improves the accuracy of a strong baseline on SParC and CoSQL, two widely used cross-domain text-to-SQL datasets. Our analysis shows that self-play simulates various conversational thematic relations, enhances cross-domain generalization and improves beam-search.

cs.CL↗

arXiv2022

UnifiedSKG: Unifying and Multi-Tasking Structured Knowledge Grounding with Text-to-Text Language Models

Structured knowledge grounding (SKG) leverages structured knowledge to complete user requests, such as semantic parsing over databases and question answering over knowledge bases. Since the inputs and outputs of SKG tasks are heterogeneous, they have been studied separately by different communities, which limits systematic and compatible research on SKG. In this paper, we overcome this limitation by proposing the UnifiedSKG framework, which unifies 21 SKG tasks into a text-to-text format, aiming to promote systematic SKG research, instead of being exclusive to a single task, domain, or dataset. We use UnifiedSKG to benchmark T5 with different sizes and show that T5, with simple modifications when necessary, achieves state-of-the-art performance on almost all of the 21 tasks. We further demonstrate that multi-task prefix-tuning improves the performance on most tasks, largely improving the overall performance. UnifiedSKG also facilitates the investigation of zero-shot and few-shot learning, and we show that T0, GPT-3, and Codex struggle in zero-shot and few-shot learning for SKG. We also use UnifiedSKG to conduct a series of controlled experiments on structured knowledge encoding variants across SKG tasks. UnifiedSKG is easily extensible to more tasks, and it is open-sourced at https://github.com/hkunlp/unifiedskg.

cs.CL↗

arXiv2022

Mask-based Latent Reconstruction for Reinforcement Learning

For deep reinforcement learning (RL) from pixels, learning effective state representations is crucial for achieving high performance. However, in practice, limited experience and high-dimensional inputs prevent effective representation learning. To address this, motivated by the success of mask-based modeling in other research fields, we introduce mask-based reconstruction to promote state representation learning in RL. Specifically, we propose a simple yet effective self-supervised method, Mask-based Latent Reconstruction (MLR), to predict complete state representations in the latent space from the observations with spatially and temporally masked pixels. MLR enables better use of context information when learning state representations to make them more informative, which facilitates the training of RL agents. Extensive experiments show that our MLR significantly improves the sample efficiency in RL and outperforms the state-of-the-art sample-efficient RL methods on multiple continuous and discrete control benchmarks. Our code is available at https://github.com/microsoft/Mask-based-Latent-Reconstruction.

cs.LG↗

arXiv2022

Radiation-free and non-Hermitian topology inertial defect states of on-chip magnons

Radiative damping is a strong dissipation source for the quantum emitters hybridized with propagating photons, electrons, or phonons, which is not easily avoidable for on-chip magnonic emitters as well that can radiate via the surface acoustic waves of the substrate. Here we demonstrate in an array of on-chip nano-magnets coupled in a long range via exchanging the surface acoustic waves that a point defect in the array, which can be introduced by the local magnon frequency shift by a local biased magnetic field or the absence of a magnetic wire, strongly localizes the magnons, in contrast to the well spreading Bloch-like collective magnon modes in such an array setting. The radiation of the magnon defect states is exponentially suppressed by the distance of the defect to the array edges. Moreover, this defect state is strikingly inertial to the non-Hermitian topology that localizes all the extended states at one boundary. Such configuration robust magnon defect states towards radiation-free limit may be suitable for high-fidelity magnon quantum information storage in the future on-chip magnonic devices.

cond-mat.mes-hall↗

arXiv2022

NL2INTERFACE: Interactive Visualization Interface Generation from Natural Language Queries

We develop NL2INTERFACE to explore the potential of generating usable interactive multi-visualization interfaces from natural language queries. With NL2INTERFACE, users can directly write natural language queries to automatically generate a fully interactive multi-visualization interface without any extra effort of learning a tool or programming language. Further, users can interact with the interfaces to easily transform the data and quickly see the results in the visualizations.

cs.HC↗

arXiv2022

A Novel Energy Efficiency Metric for Next Generation Wireless Communication Networks

As a core performance metric for green communications, the conventional energy efficiency definition has successfully resolved many issues in the energy efficient wireless network design. In the past several generations of wireless communication networks, the traditional energy efficiency measure plays an important role to guide many energy saving techniques for slow varying traffic profiles. However, for the next generation wireless networks, the traditional energy efficiency fails to capture the traffic and capacity variations of wireless networks in temporal or spatial domains, which is shown to be quite popular, especially with ultra-scale multiple antennas and space-air-ground integrated network. In this paper, we present a novel energy efficiency metric named integrated relative energy efficiency (IREE), which is able to jointly measure the traffic profiles and the network capacities from the energy efficiency perspective. On top of that, the IREE based green trade-offs have been investigated and compared with the conventional energy efficient design. Moreover, we apply the IREE based green trade-offs to evaluate several candidate technologies for 6G networks, including reconfigurable intelligent surfaces and space-air-ground integrated network. Through some analytical and numerical results, we show that the proposed IREE metric is able to capture the wireless traffic and capacity mismatch property, which is significantly different from the conventional energy efficiency metric. Since the IREE oriented design or deployment strategy is able to consider the network capacity improvement and the wireless traffic matching simultaneously, it can be regarded as a useful guidance for future energy efficient network design.

cs.NI↗

arXiv2022

Selective Annotation Makes Language Models Better Few-Shot Learners

Many recent approaches to natural language tasks are built on the remarkable abilities of large language models. Large language models can perform in-context learning, where they learn a new task from a few task demonstrations, without any parameter updates. This work examines the implications of in-context learning for the creation of datasets for new natural language tasks. Departing from recent in-context learning methods, we formulate an annotation-efficient, two-step framework: selective annotation that chooses a pool of examples to annotate from unlabeled data in advance, followed by prompt retrieval that retrieves task examples from the annotated pool at test time. Based on this framework, we propose an unsupervised, graph-based selective annotation method, voke-k, to select diverse, representative examples to annotate. Extensive experiments on 10 datasets (covering classification, commonsense reasoning, dialogue, and text/code generation) demonstrate that our selective annotation method improves the task performance by a large margin. On average, vote-k achieves a 12.9%/11.4% relative gain under an annotation budget of 18/100, as compared to randomly selecting examples to annotate. Compared to state-of-the-art supervised finetuning approaches, it yields similar performance with 10-100x less annotation cost across 10 tasks. We further analyze the effectiveness of our framework in various scenarios: language models with varying sizes, alternative selective annotation methods, and cases where there is a test data domain shift. We hope that our studies will serve as a basis for data annotations as large language models are increasingly applied to new tasks. Our code is available at https://github.com/HKUNLP/icl-selective-annotation.

cs.CL↗

arXiv2022

DC-SPP-YOLO: Dense Connection and Spatial Pyramid Pooling Based YOLO for Object Detection

Although the YOLOv2 method is extremely fast on object detection, its detection accuracy is restricted due to the low performance of its backbone network and the underutilization of multi-scale region features. Therefore, a dense connection (DC) and spatial pyramid pooling (SPP) based YOLO (DC-SPP-YOLO) method for ameliorating the object detection accuracy of YOLOv2 is proposed in this paper. Specifically, the dense connection of convolution layers is employed in the backbone network of YOLOv2 to strengthen the feature extraction and alleviate the vanishing-gradient problem. Moreover, an improved spatial pyramid pooling is introduced to pool and concatenate the multi-scale region features, so that the network can learn the object features more comprehensively. The DC-SPP-YOLO model is established and trained based on a new loss function composed of MSE (mean square error) loss and cross-entropy loss. The experimental results indicated that the mAP (mean Average Precision) of DC-SPP-YOLO is higher than that of YOLOv2 on the PASCAL VOC datasets and the UA-DETRAC datasets. The effectiveness of DC-SPP-YOLO method proposed is demonstrated.

cs.CV↗

arXiv2022

HVTR: Hybrid Volumetric-Textural Rendering for Human Avatars

We propose a novel neural rendering pipeline, Hybrid Volumetric-Textural Rendering (HVTR), which synthesizes virtual human avatars from arbitrary poses efficiently and at high quality. First, we learn to encode articulated human motions on a dense UV manifold of the human body surface. To handle complicated motions (e.g., self-occlusions), we then leverage the encoded information on the UV manifold to construct a 3D volumetric representation based on a dynamic pose-conditioned neural radiance field. While this allows us to represent 3D geometry with changing topology, volumetric rendering is computationally heavy. Hence we employ only a rough volumetric representation using a pose-conditioned downsampled neural radiance field (PD-NeRF), which we can render efficiently at low resolutions. In addition, we learn 2D textural features that are fused with rendered volumetric features in image space. The key advantage of our approach is that we can then convert the fused features into a high-resolution, high-quality avatar by a fast GAN-based textural renderer. We demonstrate that hybrid rendering enables HVTR to handle complicated motions, render high-quality avatars under user-controlled poses/shapes and even loose clothing, and most importantly, be efficient at inference time. Our experimental results also demonstrate state-of-the-art quantitative results.

cs.CV↗

arXiv2022

MCTensor: A High-Precision Deep Learning Library with Multi-Component Floating-Point

In this paper, we introduce MCTensor, a library based on PyTorch for providing general-purpose and high-precision arithmetic for DL training. MCTensor is used in the same way as PyTorch Tensor: we implement multiple basic, matrix-level computation operators and NN modules for MCTensor with identical PyTorch interface. Our algorithms achieve high precision computation and also benefits from heavily-optimized PyTorch floating-point arithmetic. We evaluate MCTensor arithmetic against PyTorch native arithmetic for a series of tasks, where models using MCTensor in float16 would match or outperform the PyTorch model with float32 or float64 precision.

cs.LG↗