arXiv subjects

Jianfeng Liu

Publications and source records attributed to Jianfeng Liu.

35 records · Page 2Linked to original sources

arXiv2023

UPRISE: Universal Prompt Retrieval for Improving Zero-Shot Evaluation

Large Language Models (LLMs) are popular for their impressive abilities, but the need for model-specific fine-tuning or task-specific prompt engineering can hinder their generalization. We propose UPRISE (Universal Prompt Retrieval for Improving zero-Shot Evaluation), which tunes a lightweight and versatile retriever that automatically retrieves prompts for a given zero-shot task input. Specifically, we demonstrate universality in a cross-task and cross-model scenario: the retriever is tuned on a diverse set of tasks, but tested on unseen task types; we use a small frozen LLM, GPT-Neo-2.7B, for tuning the retriever, but test the retriever on different LLMs of much larger scales, such as BLOOM-7.1B, OPT-66B and GPT3-175B. Additionally, we show that UPRISE mitigates the hallucination problem in our experiments with ChatGPT, suggesting its potential to improve even the strongest LLMs. Our model and code are available at https://github.com/microsoft/LMOps.

cs.CL↗

arXiv2023

Visual Instruction Tuning with Polite Flamingo

Recent research has demonstrated that the multi-task fine-tuning of multi-modal Large Language Models (LLMs) using an assortment of annotated downstream vision-language datasets significantly enhances their performance. Yet, during this process, a side effect, which we termed as the "multi-modal alignment tax", surfaces. This side effect negatively impacts the model's ability to format responses appropriately -- for instance, its "politeness" -- due to the overly succinct and unformatted nature of raw annotations, resulting in reduced human preference. In this paper, we introduce Polite Flamingo, a multi-modal response rewriter that transforms raw annotations into a more appealing, "polite" format. Polite Flamingo is trained to reconstruct high-quality responses from their automatically distorted counterparts and is subsequently applied to a vast array of vision-language datasets for response rewriting. After rigorous filtering, we generate the PF-1M dataset and further validate its value by fine-tuning a multi-modal LLM with it. Combined with novel methodologies including U-shaped multi-stage tuning and multi-turn augmentation, the resulting model, Clever Flamingo, demonstrates its advantages in both multi-modal understanding and response politeness according to automated and human evaluations.

cs.CV↗

arXiv2023

Orca: A Few-shot Benchmark for Chinese Conversational Machine Reading Comprehension

The conversational machine reading comprehension (CMRC) task aims to answer questions in conversations, which has been a hot research topic in recent years because of its wide applications. However, existing CMRC benchmarks in which each conversation is assigned a static passage are inconsistent with real scenarios. Thus, model's comprehension ability towards real scenarios are hard to evaluate reasonably. To this end, we propose the first Chinese CMRC benchmark Orca and further provide zero-shot/few-shot settings to evaluate model's generalization ability towards diverse domains. We collect 831 hot-topic driven conversations with 4,742 turns in total. Each turn of a conversation is assigned with a response-related passage, aiming to evaluate model's comprehension ability more reasonably. The topics of conversations are collected from social media platform and cover 33 domains, trying to be consistent with real scenarios. Importantly, answers in Orca are all well-annotated natural responses rather than the specific spans or short phrase in previous datasets. Besides, we implement three strong baselines to tackle the challenge in Orca. The results indicate the great challenge of our CMRC benchmark. Our datatset and checkpoints are available at https://github.com/nuochenpku/Orca.

cs.CL↗

arXiv2023

Semiquantum key distribution using initial states in only one basis without the classical user measuring

From the perspective of resource theory, it is interesting to achieve the same quantum task using as few quantum resources as possible. Semiquantum key distribution (SQKD), which allows a quantum user to share a confidential key with a classical user who prepares and operates qubits in only one basis, is an important example for studying this issue. To further limit the quantum resources used by users, in this paper, we constructed the first SQKD protocol which restricts the quantum user to prepare quantum states in only one basis and removes the classical user's measurement capability. Furthermore, we prove that the constructed protocol is unconditionally secure by deriving a key rate expression of the error rate in the asymptotic scenario. The work of this paper provides inspiration for achieving quantum superiority with minimal quantum resources.

quant-ph↗

arXiv2023

Application of Causal Inference Techniques to the Maximum Weight Independent Set Problem

A powerful technique for solving combinatorial optimization problems is to reduce the search space without compromising the solution quality by exploring intrinsic mathematical properties of the problems. For the maximum weight independent set (MWIS) problem, using an upper bound lemma which says the weight of any independent set not contained in the MWIS is bounded from above by the weight of the intersection of its closed neighbor set and the MWIS, we give two extension theorems -- independent set extension theorem and vertex cover extension theorem. With them at our disposal, two types of causal inference techniques (CITs) are proposed on the assumption that a vertex is strongly reducible (included or not included in all MWISs) or reducible (contained or not contained in a MWIS). One is a strongly reducible state-preserving technique, which extends a strongly reducible vertex into a vertex set where all vertices have the same strong reducibility. The other, as a reducible state-preserving technique, extends a reducible vertex into a vertex set with the same reducibility as that vertex and creates some weighted packing constraints to narrow the search space. Numerical experiments show that our CITs can help reduction algorithms find much smaller remaining graphs, improve the ability of exact algorithms to find the optimal solutions and help heuristic algorithms produce approximate solutions of better quality. In particular, detailed tests on $12$ representative graphs generated from datasets in Network Data Repository demonstrate that, compared to the state-of-the-art algorithms, the size of remaining graphs is further reduced by more than 32.6%, and the number of solvable instances is increased from 1 to 5.

math.OC↗

arXiv2022

Exploring Contextual Relationships for Cervical Abnormal Cell Detection

Cervical abnormal cell detection is a challenging task as the morphological discrepancies between abnormal and normal cells are usually subtle. To determine whether a cervical cell is normal or abnormal, cytopathologists always take surrounding cells as references to identify its abnormality. To mimic these behaviors, we propose to explore contextual relationships to boost the performance of cervical abnormal cell detection. Specifically, both contextual relationships between cells and cell-to-global images are exploited to enhance features of each region of interest (RoI) proposals. Accordingly, two modules, dubbed as RoI-relationship attention module (RRAM) and global RoI attention module (GRAM), are developed and their combination strategies are also investigated. We establish a strong baseline by using Double-Head Faster R-CNN with feature pyramid network (FPN) and integrate our RRAM and GRAM into it to validate the effectiveness of the proposed modules. Experiments conducted on a large cervical cell detection dataset reveal that the introduction of RRAM and GRAM both achieves better average precision (AP) than the baseline methods. Moreover, when cascading RRAM and GRAM, our method outperforms the state-of-the-art (SOTA) methods. Furthermore, we also show the proposed feature enhancing scheme can facilitate both image-level and smear-level classification. The code and trained models are publicly available at https://github.com/CVIU-CSU/CR4CACD.

cs.CV↗

arXiv2020

GoChat: Goal-oriented Chatbots with Hierarchical Reinforcement Learning

A chatbot that converses like a human should be goal-oriented (i.e., be purposeful in conversation), which is beyond language generation. However, existing dialogue systems often heavily rely on cumbersome hand-crafted rules or costly labelled datasets to reach the goals. In this paper, we propose Goal-oriented Chatbots (GoChat), a framework for end-to-end training chatbots to maximize the longterm return from offline multi-turn dialogue datasets. Our framework utilizes hierarchical reinforcement learning (HRL), where the high-level policy guides the conversation towards the final goal by determining some sub-goals, and the low-level policy fulfills the sub-goals by generating the corresponding utterance for response. In our experiments on a real-world dialogue dataset for anti-fraud in financial, our approach outperforms previous methods on both the quality of response generation as well as the success rate of accomplishing the goal.

cs.CL↗

arXiv2020

Applying Cyclical Learning Rate to Neural Machine Translation

In training deep learning networks, the optimizer and related learning rate are often used without much thought or with minimal tuning, even though it is crucial in ensuring a fast convergence to a good quality minimum of the loss function that can also generalize well on the test dataset. Drawing inspiration from the successful application of cyclical learning rate policy for computer vision related convolutional networks and datasets, we explore how cyclical learning rate can be applied to train transformer-based neural networks for neural machine translation. From our carefully designed experiments, we show that the choice of optimizers and the associated cyclical learning rate policy can have a significant impact on the performance. In addition, we establish guidelines when applying cyclical learning rates to neural machine translation tasks. Thus with our work, we hope to raise awareness of the importance of selecting the right optimizers and the accompanying learning rate policy, at the same time, encourage further research into easy-to-use learning rate policies.

cs.LG↗

arXiv2019

Prototype of a front-end readout ASIC designed for the Water Cherenkov Detector Array in LHAASO

The Large High Altitude Air Shower Observatory is in the R&D phase, in which the Water Cherenkov Detector Array is an important part. The signals of Photo-Multiplier Tubes would vary from single photo electron to 4000 photo electrons, and both high precision charge and time measurement is required. To simplify the signal processing chain, the charge-to-time conversion method is employed. A prototype of the front-end readout ASIC is designed and fabricated in Chartered 0.35 μm CMOS technology, which integrates time disctrimination and converts the input charge information to pulse widths. With Time-to-Digital Converters, both time and charge can be digitized at the same time. We have conducted initial tests on this chip, and the results indicate that a time resolution better than 0.5 ns is achieved over the full dynamic range (1~ 4000 photo electrons, corresponding to 0.75 ~ 3000 pC with the threshold of 0.188 pC); the charge resolution is better than 1% with large input amplitudes (500 ~ 4000 photo electrons), and remains better than 15% with a 1 photo electron input amplitude, which is beyond the application requirement.

physics.ins-det↗

arXiv2018

Efficient Misalignment-Robust Multi-Focus Microscopical Images Fusion

In this paper we propose a very efficient method to fuse the unregistered multi-focus microscopical images based on the speed-up robust features (SURF). Our method follows the pipeline of first registration and then fusion. However, instead of treating the registration and fusion as two completely independent stage, we propose to reuse the determinant of the approximate Hessian generated in SURF detection stage as the corresponding salient response for the final image fusion, thus it enables nearly cost-free saliency map generation. In addition, due to the adoption of SURF scale space representation, our method can generate scale-invariant saliency map which is desired for scale-invariant image fusion. We present an extensive evaluation on the dataset consisting of several groups of unregistered multi-focus 4K ultra HD microscopic images with size of 4112 x 3008. Compared with the state-of-the-art multi-focus image fusion methods, our method is much faster and achieve better results in the visual performance. Our method provides a flexible and efficient way to integrate complementary and redundant information from multiple multi-focus ultra HD unregistered images into a fused image that contains better description than any of the individual input images. Code is available at https://github.com/yiqingmy/JointRF.

cs.CV↗

arXiv2018

Scale-Invariant Structure Saliency Selection for Fast Image Fusion

In this paper, we present a fast yet effective method for pixel-level scale-invariant image fusion in spatial domain based on the scale-space theory. Specifically, we propose a scale-invariant structure saliency selection scheme based on the difference-of-Gaussian (DoG) pyramid of images to build the weights or activity map. Due to the scale-invariant structure saliency selection, our method can keep both details of small size objects and the integrity information of large size objects in images. In addition, our method is very efficient since there are no complex operation involved and easy to be implemented and therefore can be used for fast high resolution images fusion. Experimental results demonstrate the proposed method yields competitive or even better results comparing to state-of-the-art image fusion methods both in terms of visual quality and objective evaluation metrics. Furthermore, the proposed method is very fast and can be used to fuse the high resolution images in real-time. Code is available at https://github.com/yiqingmy/Fusion.

cs.CV↗

arXiv2018

Global Solution Strategies for the Network-Constrained Unit Commitment Problem with AC Transmission Constraints

We propose a novel global solution algorithm for the network-constrained unit commitment problem incorporating a nonlinear alternating current model of the transmission network, which is a nonconvex mixed-integer nonlinear programming (MINLP) problem. Our algorithm is based on the multi-tree global optimization methodology, which iterates between a mixed-integer lower-bounding problem and a nonlinear upper-bounding problem. We exploit the mathematical structure of the unit commitment problem with AC power flow constraints (UC-AC) and leverage optimization-based bounds tightening, second-order cone relaxations, and piecewise outer approximations to guarantee a globally optimal solution at convergence. Numerical results on four benchmark problems illustrate the effectiveness of our algorithm, both in terms of convergence rate and solution quality.

math.OC↗

arXiv2015

An Analogue Front-End ASIC Prototype Designed For PMT Signal Readout

The Large High Altitude Air Shower Observatory (LHAASO) is designed for high energy gamma ray and cosmic ray detection. A Water Cherenkov Detector Array which is sensitive to gamma ray showers above a few hundred GeV is proposed to survey gamma ray sources. The WCDA consists of 3600 PhotoMultiplier Tubes (PMT) which collect the Cherenkov light produced by the shower particles in water. Both high precision time and charge measurement are required over a large dynamic range from 1 photo electron (P.E.) to 4000 P.E. Prototype of an analogue front-end Application Specific Integrated Circuit (ASIC) fabricated in Chartered 0.35 μm CMOS technology is designed to read out PMT signal in the WCDA. This ASIC employs leading edge discrimination and RC4 shaping structure; combined with the following Time-to-Digital Converter (TDC) and Analog-to-Digital Converter (ADC), both the arrival time and charge of the PMT signal can be measured. Initial test results indicate that time resolution is better than 350 ps and charge resolution is better than 10% at 1 P.E. and better than 1% with large input signals (300 P.E. to 4000 P.E.). Besides, this ASIC has a small channel-to-channel crosstalk and low ambient temperature dependency.

physics.ins-det↗

arXiv2015

Evaluation of a front-end ASIC for the readout of PMTs in large dynamic range

The Large High Altitude Air Shower Observatory (LHAASO) project has been proposed for the survey and study of cosmic rays. In the LHAASO project, the Water Cherenkov Detector Array (WCDA) is one of major detectors for searching gamma ray sources. A Charge-to-Time Convertor (QTC) ASIC (Application Specification Integrated Circuit) fabricated in Global Foundry 0.35 μm CMOS technology, has been developed for readout of Photomultiplier Tubes (PMTs) in the WCDA. This paper focuses on the evaluation of this front-end readout ASIC performance. Test results indicate that the time resolution is better than 400 ps and the charge resolution is better than 1% with large input signals and remains better than 15% @ 1 Photo Electron (P.E.), both beyond the application requirement. Moreover, this ASIC has a weak ambient temperature dependence, low input rate dependence and high channel-to-channel isolation.

physics.ins-det↗

arXiv2008

Dominant role of GABAB2 and Gbetagamma for GABAB receptor-mediated-ERK1/2/CREB pathway in cerebellar neurons

gamma-aminobutyric acid type B (GABA(B)) receptor is an allosteric complex made of two subunits, GABA(B1) and GABA(B2). GABA(B2) plays a major role in the coupling to G protein whereas GABA(B1) binds GABA. It has been shown that GABA(B) receptor activates ERK(1/2) in neurons of the central nervous system, but the molecular mechanisms underlying this event are poorly characterized. Here, we demonstrate that activation of GABA(B) receptor by either GABA or the selective agonist baclofen induces ERK(1/2) phosphorylation in cultured cerebellar granule neurons. We also show that CGP7930, a positive allosteric regulator specific of GABA(B2), alone can induce the phosphorylation of ERK(1/2). PTX, a G(i/o) inhibitor, abolishes both baclofen and CGP7930-mediated-ERK(1/2) phosphorylation. Moreover, both baclofen and CGP7930 induce ERK-dependent CREB phosphorylation. Furthermore, by using LY294002, a PI-3 kinase inhibitor, and a C-term of GRK-2 that has been reported to sequester Gbetagamma subunits, we demonstrate the role of Gbetagamma in GABA(B) receptor-mediated-ERK(1/2) phosphorylation. In conclusion, the activation of GABA(B) receptor leads to ERK(1/2) phosphorylation via the coupling of GABA(B2) to G(i/o) and by releasing Gbetagamma subunits which in turn induce the activation of CREB. These findings suggest a role of GABA(B) receptor in long-term change in the central nervous system.

q-bio.NC↗

arXiv2008

Functioning of the dimeric GABA(B) receptor extracellular domain revealed by glycan wedge scanning

The G-protein-coupled receptor (GPCR) activated by the neurotransmitter GABA is made up of two subunits, GABA(B1) and GABA(B2). GABA(B1) binds agonists, whereas GABA(B2) is required for trafficking GABA(B1) to the cell surface, increasing agonist affinity to GABA(B1), and activating associated G proteins. These subunits each comprise two domains, a Venus flytrap domain (VFT) and a heptahelical transmembrane domain (7TM). How agonist binding to the GABA(B1) VFT leads to GABA(B2) 7TM activation remains unknown. Here, we used a glycan wedge scanning approach to investigate how the GABA(B) VFT dimer controls receptor activity. We first identified the dimerization interface using a bioinformatics approach and then showed that introducing an N-glycan at this interface prevents the association of the two subunits and abolishes all activities of GABA(B2), including agonist activation of the G protein. We also identified a second region in the VFT where insertion of an N-glycan does not prevent dimerization, but blocks agonist activation of the receptor. These data provide new insight into the function of this prototypical GPCR and demonstrate that a change in the dimerization interface is required for receptor activation.

q-bio.NC↗

arXiv2008

Molecular determinants involved in the allosteric control of agonist affinity in the GABAB receptor by the GABAB2 subunit

The gamma-aminobutyric acid type B (GABAB) receptor is an allosteric complex made of two subunits, GABAB1 (GB1) and GABAB2 (GB2). Both subunits are composed of an extracellular Venus flytrap domain (VFT) and a heptahelical domain (HD). GB1 binds GABA, and GB2 plays a major role in G-protein activation as well as in the high agonist affinity state of GB1. How agonist affinity in GB1 is regulated in the receptor remains unknown. Here, we demonstrate that GB2 VFT is a major molecular determinant involved in this control. We show that isolated versions of GB1 and GB2 VFTs in the absence of the HD and C-terminal tail can form hetero-oligomers as shown by time-resolved fluorescence resonance energy transfer (based on HTRF technology). GB2 VFT and its association with GB1 VFT controlled agonist affinity in GB1 in two ways. First, GB2 VFT exerted a direct action on GB1 VFT, as it slightly increased agonist affinity in isolated GB1 VFT. Second and most importantly, GB2 VFT prevented inhibitory interaction between the two main domains (VFT and HD) of GB1. According to this model, we propose that GB1 HD prevents the possible natural closure of GB1 VFT. In contrast, GB2 VFT facilitates this closure. Finally, such inhibitory contacts between HD and VFT in GB1 could be similar to those important to maintain the inactive state of the receptor.

q-bio.BM↗