Filters








8,636 Hits in 4.7 sec

Zero-Shot Task Generalization with Multi-Task Deep Reinforcement Learning [article]

Junhyuk Oh, Satinder Singh, Honglak Lee, Pushmeet Kohli
2017 arXiv   pre-print
As a step towards developing zero-shot task generalization capabilities in reinforcement learning (RL), we introduce a new RL problem where the agent should learn to execute sequences of instructions after  ...  To deal with delayed reward, we propose a new neural architecture in the meta controller that learns when to update the subtask, which makes learning more efficient.  ...  Zero-Shot Task Generalization. There have been a few papers on zero-shot generalization to new tasks.  ... 
arXiv:1706.05064v2 fatcat:zfbczd37bja2xkcthivo6lvsem

A New Approach for Training Cobots from Small Amount of Data in Industry 5.0

Khalid Jabrane, Mohammed Bousmah
2021 International Journal of Advanced Computer Science and Applications  
Many AI techniques are needed, ranging from visual processing to symbolic reasoning, task planning to mind building theory, reactive control to action recognition and learning.  ...  All Artificial Intelligence learning techniques require the training of algorithms with huge data. Collecting and storing this data takes time and requires increasing computer memory.  ...  These functions will be implemented by three Deep Neural Network called:  ZSL-DNN (Zero-Shot Learning Deep Neural Network).  FSL-DNN (Few-Shot Learning Deep Neural Network).  MTDRL-DNN (Multi-Task Deep  ... 
doi:10.14569/ijacsa.2021.0121070 fatcat:x46lmt7c7jacvge2bcmw7nelwq

Zero-Shot Terrain Generalization for Visual Locomotion Policies [article]

Alejandro Escontrela, George Yu, Peng Xu, Atil Iscen, Jie Tan
2020 arXiv   pre-print
We frame this challenge as a multi-task reinforcement learning problem and define each task as a type of terrain that the robot needs to traverse.  ...  As a result, the learned controller demonstrates excellent zero-shot generalization capabilities and can navigate 13 different environments, including stairs, rugged land, cluttered offices, and indoor  ...  Multi-Task Reinforcement Learning Multi-task reinforcement learning (MTRL) [20] is a promising approach to train generalizable policies that can accomplish a wide variety of tasks. Hessel et. al.  ... 
arXiv:2011.05513v1 fatcat:vz2aa4742fepdat2deifangalq

Zero-Shot Visual Slot Filling as Question Answering [article]

Larry Heck, Simon Heck
2022 arXiv   pre-print
An approach to further refine the model with multi-task training is presented.  ...  The multi-task approach facilitates the incorporation of a large number of successive refinements and transfer learning across tasks.  ...  Our new Visual Slots as QA approach shows an improved F1 score of 0.52 with zero-shot multi-task training for the Visual Slot dataset.  ... 
arXiv:2011.12340v2 fatcat:wzfvi57j5bedzlmm4qhbpx2oba

Compositional Multi-Object Reinforcement Learning with Linear Relation Networks [article]

Davide Mambelli, Frederik Träuble, Stefan Bauer, Bernhard Schölkopf, Francesco Locatello
2022 arXiv   pre-print
In this paper, we focus on models that can learn manipulation tasks in fixed multi-object settings and extrapolate this skill zero-shot without any drop in performance when the number of objects changes  ...  Although reinforcement learning has seen remarkable progress over the last years, solving robust dexterous object-manipulation tasks in multi-object settings remains a challenge.  ...  Can we train RL agents that learn such manipulation tasks with a fixed multi-object setting -training with only two distractors -and extrapolate this skill zero-shot when the number of distractors changes  ... 
arXiv:2201.13388v1 fatcat:4rn2ioo6dngkfpqfonp52duolu

Learning Modular Neural Network Policies for Multi-Task and Multi-Robot Transfer [article]

Coline Devin, Abhishek Gupta, Trevor Darrell, Pieter Abbeel, Sergey Levine
2016 arXiv   pre-print
Using a novel neural network architecture, we demonstrate the effectiveness of our transfer method for enabling zero-shot generalization with a variety of robots and tasks in simulation for both visual  ...  Using deep reinforcement learning to train general purpose neural network policies alleviates some of the burden of manual representation engineering by using expressive policy classes, but exacerbates  ...  DISCUSSION AND FUTURE WORK In this paper, we presented modular policy networks, a method for enabling multi-robot and multi-task transfer with reinforcement learning.  ... 
arXiv:1609.07088v1 fatcat:dq57nzd5ajafhbjqcydpdhnlou

Continuous Coordination As a Realistic Scenario for Lifelong Learning [article]

Hadi Nekoei, Akilesh Badrinaaraayanan, Aaron Courville, Sarath Chandar
2021 arXiv   pre-print
Current deep reinforcement learning (RL) algorithms are still highly task-specific and lack the ability to generalize to new environments.  ...  In this work, we introduce a multi-agent lifelong learning testbed that supports both zero-shot and few-shot settings.  ...  also grateful to Hengyuan Hu for patiently answering queries regarding the Hanabi SAD code repository, Darshan Patil and Rodrigo Chavez Zavaleta for reviewing our code and Olexa Bilaniuk for helping us with  ... 
arXiv:2103.03216v2 fatcat:czld45ilrjfaxldxzmzcndfy5e

Multi Task Learning For Zero Shot Performance Prediction of Multilingual Models [article]

Kabir Ahuja, Shanu Kumar, Sandipan Dandapat, Monojit Choudhury
2022 arXiv   pre-print
In this work, we build upon some of the existing techniques for predicting the zero-shot performance on a task, by modeling it as a multi-task learning problem.  ...  Our approach also lends us the ability to perform a much more robust feature selection and identify a common set of features that influence zero-shot performance across a variety of tasks.  ...  And fourth, our multi-task framework in general lends us with a much more robust selection of features affecting the zero-shot performance.  ... 
arXiv:2205.06130v1 fatcat:zirvmdhtwnec5kkvsp2elxsose

Encoding formulas as deep networks: Reinforcement learning for zero-shot execution of LTL formulas [article]

Yen-Ling Kuo, Boris Katz, Andrei Barbu
2020 arXiv   pre-print
This is a novel form of multi-task learning for RL agents where agents learn from one diverse set of tasks and generalize to a new set of diverse tasks.  ...  The input LTL formulas have never been seen before, yet the network performs zero-shot generalization to satisfy them.  ...  While we only discuss LTL here, this approach suggests how other logics might similarly be encoded to create new powerful zero-shot deep approaches to reinforcement learning.  ... 
arXiv:2006.01110v2 fatcat:5oewaleiqzfubm2wuue6uagb4m

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks [article]

Chelsea Finn, Pieter Abbeel, Sergey Levine
2017 arXiv   pre-print
gradient reinforcement learning with neural network policies.  ...  In our approach, the parameters of the model are explicitly trained such that a small number of gradient steps with a small amount of training data from a new task will produce good generalization performance  ...  gradient reinforcement learning, with minimal modification.  ... 
arXiv:1703.03400v3 fatcat:c2f3ayn6kretpb6vga7hrziwoi

Generalization in Dexterous Manipulation via Geometry-Aware Multi-Task Learning [article]

Wenlong Huang, Igor Mordatch, Pieter Abbeel, Deepak Pathak
2021 arXiv   pre-print
In this work, we show that policies learned by existing reinforcement learning algorithms can in fact be generalist when combined with multi-task learning and a well-chosen object representation.  ...  Interestingly, we find that multi-task learning with object point cloud representations not only generalizes better but even outperforms the single-object specialist policies on both training as well as  ...  The work was supported in part by Berkeley Deep Drive, NSF IIS-2024594 and GoodAI Research Award.  ... 
arXiv:2111.03062v1 fatcat:5sm22x4mbbbpbapw5wj2mwpgdm

CRNet: Cross-Reference Networks for Few-Shot Segmentation [article]

Weide Liu, Chi Zhang, Guosheng Lin, Fayao Liu
2020 arXiv   pre-print
Recently, few-shot segmentation is proposed to solve this problem. Few-shot segmentation aims to learn a segmentation model that can be generalized to novel classes with only a few training images.  ...  With a cross-reference mechanism, our network can better find the co-occurrent objects in the two images, thus helping the few-shot segmentation task.  ...  This research is also partly supported by the Delta-NTU Corporate Lab with funding support from Delta Electronics Inc. and the National Research Foundation (NRF) Singapore.  ... 
arXiv:2003.10658v1 fatcat:vuyhq3h57rakteroednjwpd2jy

SAFFNet: Self-Attention-Based Feature Fusion Network for Remote Sensing Few-Shot Scene Classification

Joseph Kim, Mingmin Chi
2021 Remote Sensing  
Here, the feature weighting value can be fine-tuned by the support set in the few-shot learning task.  ...  To solve this problem, few-shot learning methods are usually adopted to recognize new categories with only a few (out-of-bag) labeled samples together with the known classes available in the (large-scale  ...  Comparison of traditional supervised, zero-shot and few-shot learning for classification tasks.  ... 
doi:10.3390/rs13132532 fatcat:fonmvysiczct5kr7wwl3xe2xdm

Unicorn: Continual Learning with a Universal, Off-policy Agent [article]

Daniel J. Mankowitz, Augustin Žídek, André Barreto, Dan Horgan, Matteo Hessel, John Quan, Junhyuk Oh, Hado van Hasselt, David Silver, Tom Schaul
2018 arXiv   pre-print
To test continual learning capabilities we consider a challenging 3D domain with an implicit sequence of tasks and sparse rewards.  ...  Instead, some tasks continually grow in complexity, in tandem with the agent's competence.  ...  to solve related tasks, even with zero-shot transfer and (3) solving tasks with deep dependencies.  ... 
arXiv:1802.08294v2 fatcat:pmxidnqirngjtp6oyxrcrmbv64

Toward Open-World Electroencephalogram Decoding Via Deep Learning: A Comprehensive Survey [article]

Xun Chen, Chang Li, Aiping Liu, Martin J. McKeown, Ruobing Qian, Z. Jane Wang
2021 arXiv   pre-print
In recent years, deep learning (DL) has emerged as a potential solution for such problems due to its superior capacity in feature extraction.  ...  Combining DL with domain-specific knowledge may allow for development of robust approaches to decode brain activity even with small-sample data.  ...  Her research interests include statistical signal processing and machine learning, with applications in digital media and biomedical data analytics.  ... 
arXiv:2112.06654v2 fatcat:roxf5k7ypfcvtdzz3pbho3kdri
« Previous Showing results 1 — 15 out of 8,636 results