Tao Meng
@TaoMeng10
Followers
179
Following
40
Media
2
Statuses
22
PhD@UCLANLP, NLP&ML research focus: injecting constraints in NLP/ML models https://t.co/mjHcc1ouiV
Los Angeles
Joined May 2019
Happy to share our work on Controllable Text Generation with NeurAlly-Decomposed Oracle (NADO) accepted at #Neurips2022! Looks like our model is excited about NeurIPS as much as we do. See more in
3
22
102
RT @HonghuaZhang2: Reliable control of large language models is a crucial problem. We propose GeLaTo (Generating Language with Tractable Co….
arxiv.org
Despite the success of autoregressive large language models in text generation, it remains a major challenge to generate text that satisfies complex constraints: sampling from the conditional...
0
38
0
Blazing-fast image creation – using just your voice. Try Grok Imagine.
284
557
3K
RT @kaiwei_chang: ACL/ICML highlights threats of LMs like ChatGPT generating paper content. However, I'm more concerned about reviews. Exam….
0
22
0
Welcome to our controllable text generation poster this afternoon!.
Dear friends at #NeurIPS2022, #PlusLab_UCLANLP will be presenting three papers at the main conference and one at the ENLSP workshop. This is a thread of some details about our presentations. My students and I are looking forward to seeing you at NOLA! (1/5).
0
0
3
RT @sidilu_pluslab: It's an honor of mine to share our work InsNet! We propose a unified, efficient, and powerful framework for both sequen….
0
5
0
RT @VioletNPeng: I’m super excited to share several fresh off the press works on controllable (creative) generation this Thursday at Stanfo….
0
12
0
The code for reproduce the results is released at Special thanks to my collaborators Sidi Lu, Nanyun Peng and Kai-Wei Chang! @sidilu_pluslab @VioletNPeng @kaiwei_chang.
github.com
Constrained Decoding Project. Contribute to MtSomeThree/constrDecoding development by creating an account on GitHub.
0
1
0
We conduct experiments on two applications: (1) text generation with lexical constraints and (2) machine translation with formality control, demonstrating our framework guides the base model towards the given oracles(both rule and classifier-based) while maintaining high quality.
1
1
0
The token-level guidance is approximated by a neural model trained with examples sampled from the base model. We present the closed-form optimal solution to incorporating the token-level guidance into the base model. We further provide a theoretical analysis of the approximation.
1
1
1
How to control the models to satisfy pre-defined sentence-level attributes is an open challenge. Given a pre-trained language model and a sequence-level boolean oracle function, we propose to decompose the oracle into token-level guidance to steer the base model in generation.
1
1
0
RT @LiLiunian: Happy to share our work on Object Detection in the Wild through Grounded Language Image Pre-training (GLIP) (Oral at #CVPR20….
0
41
0
RT @HonghuaZhang2: Can language models learn to reason by end-to-end training? We show that near-perfect test accuracy is deceiving: instea….
arxiv.org
Logical reasoning is needed in a wide range of NLP tasks. Can a BERT model be trained end-to-end to solve logical reasoning problems presented in natural language? We attempt to answer this...
0
58
0
RT @uclanlp: UCLA Chang's (@kaiwei_chang ) and Plus lab (@VioletNPeng) will present papers and a tutorial on topics including Fairness & Ro….
0
29
0
RT @natarajan_prem: Thrilled to participate in the launch event of the Science Hub at UCLA. Looking forward to the many advances that this….
0
9
0
Besides, we provide a theoretical analysis about the tightness of the polytopes and the reliability of the mined constraints. Paper/Code: .4/4.
web.cs.ucla.edu
An Integer Linear Programming Framework for Mining Constraints from Data Share this page: An Integer Linear Programming Framework for Mining Constraints from Data Tao Meng and Kai-Wei Chang, in ICML,...
0
0
1
We show that our approach learns to solve 9x9 Sudoku and minimal spanning tree problems from examples without giving underlying rules. It can also integrate with a neural net to learn the hierarchical label structure of a multi-label classification task 3/4.
1
0
2
We present an integer linear programming (ILP) framework for constraint mining. We formulate inference of structured labels as ILP. Given coefficients of obj. function and corresponding solution, we mine constraints by estimating the outer&inner polytopes of the feasible set 2/4
1
0
2
Can a model learn problem structure/constraints from data? E.g., given pairs of adjacent matrix and corresponding minimal spanning tree, can a model learn to solve MST? Check out our #ICML2021 paper on constraint mining with ILP w/@kaiwei_chang 1/4.
web.cs.ucla.edu
An Integer Linear Programming Framework for Mining Constraints from Data Share this page: An Integer Linear Programming Framework for Mining Constraints from Data Tao Meng and Kai-Wei Chang, in ICML,...
1
5
24
RT @kaiwei_chang: UCLA-NLP will present at #acl2020nlp. 👇Check out our ACL/TACL papers and pre-recorded talks at.Lo….
0
9
0
RT @kaiwei_chang: The problem is much more beyond data bias. ML systems can be poorly calibrated ( and bias in the….
0
48
0