Paper List

The following papers were announced on the learn2learn Twitter account. You can submit unannounced and meta-learning related papers through the following Google Form. (It does not matter if they are old or new, but they shouldn't be already announced.)


Announce any paper via the Google Form to announce papers, also available below.

Submitted Papers

Grounded Language Learning Fast and Slow
by Hill, Felix and Tieleman, Olivier and von Glehn, Tamara and Wong, Nathaniel and Merzic, Hamza and Clark, Stephen

Sparse Meta Networks for Sequential Adaptation and its Application to Adaptive Language Modelling
by Munkhdalai, Tsendsuren

Learning with Differentiable Perturbed Optimizers
by Berthet, Quentin and Blondel, Mathieu and Teboul, Olivier and Cuturi, Marco and Vert, Jean-Philippe and Bach, Francis

What is being transferred in transfer learning?
by Neyshabur, Behnam and Sedghi, Hanie and Zhang, Chiyuan

On modulating the gradient for meta-learning
by Simon, Christian and Koniusz, Piotr and Nock, Richard and Harandi, Mehrtash

Meta-Learning with Shared Amortized Variational Inference
by Iakovleva, Ekaterina and Verbeek, Jakob and Alahari, Karteek

learn2learn: A Library for Meta-Learning Research
by Arnold, Sébastien M R and Mahajan, Praateek and Datta, Debajyoti and Bunner, Ian and Zarkias, Konstantinos Saitas

A Universal Representation Transformer Layer for Few-Shot Image Classification
by Liu, Lu and Hamilton, William and Long, Guodong and Jiang, Jing and Larochelle, Hugo

Safe Model-Based Meta-Reinforcement Learning: A Sequential Exploration-Exploitation Framework
by Lew, Thomas and Sharma, Apoorva and Harrison, James and Pavone, Marco

Learning to Learn in a Semi-Supervised Fashion
by Chen, Yun-Chun and Chou, Chao-Te and Wang, Yu-Chiang Frank

The Advantage of Conditional Meta-Learning for Biased Regularization and Fine-Tuning
by Denevi, Giulia and Pontil, Massimiliano and Ciliberto, Carlo

Adaptive Multi-level Hyper-gradient Descent
by Jie, Renlong and Gao, Junbin and Vasnev, Andrey and Tran, Minh-Ngoc

Few-Shot Image Classification via Contrastive Self-Supervised Learning
by Li, Jianyi and Liu, Guizhong

Does MAML really want feature reuse only?
by Oh, Jaehoon and Yoo, Hyungjun and Kim, Changhwan and Yun, Se-Young

Meta Learning MPC using Finite-Dimensional Gaussian Process Approximations
by Arcari, Elena and Carron, Andrea and Zeilinger, Melanie N

Offline Meta-Reinforcement Learning with Advantage Weighting
by Mitchell, Eric and Rafailov, Rafael and Peng, Xue Bin and Levine, Sergey and Finn, Chelsea

Explore then Execute: Adapting without Rewards via Factorized Meta-Reinforcement Learning
by Liu, Evan Zheran and Raghunathan, Aditi and Liang, Percy and Finn, Chelsea

Offline Meta Reinforcement Learning
by Dorfman, Ron and Tamar, Aviv

Few-Shot Learning via Learning the Representation, Provably
by Du, Simon S and Hu, Wei and Kakade, Sham M and Lee, Jason D and Lei, Qi

Multi-Task Reinforcement Learning as a Hidden-Parameter Block MDP
by Zhang, Amy and Sodhani, Shagun and Khetarpal, Khimya and Pineau, Joelle

CAMPs: Learning Context-Specific Abstractions for Efficient Planning in Factored MDPs
by Chitnis, Rohan and Silver, Tom and Kim, Beomjoon and Kaelbling, Leslie Pack and Lozano-Perez, Tomas

Unsupervised Learning of Visual Features by Contrasting Cluster Assignments
by Caron, Mathilde and Misra, Ishan and Mairal, Julien and Goyal, Priya and Bojanowski, Piotr and Joulin, Armand

MiCo: Mixup Co-Training for Semi-Supervised Domain Adaptation
by Yang, Luyu and Wang, Yan and Gao, Mingfei and Shrivastava, Abhinav and Weinberger, Kilian Q and Chao, Wei-Lun and Lim, Ser-Nam

Adaptive Task Sampling for Meta-Learning
by Liu, Chenghao and Wang, Zhihao and Sahoo, Doyen and Fang, Yuan and Zhang, Kun and Hoi, Steven C H

Discovering Reinforcement Learning Algorithms
by Oh, Junhyuk and Hessel, Matteo and Czarnecki, Wojciech M and Xu, Zhongwen and van Hasselt, Hado and Singh, Satinder and Silver, David

On the Outsized Importance of Learning Rates in Local Update Methods
by Charles, Zachary and Kone{\v c}n{\'y}, Jakub

Global Convergence and Induced Kernels of Gradient-Based Meta-Learning with Neural Nets
by Wang, Haoxiang and Sun, Ruoyu and Li, Bo

On the Iteration Complexity of Hypergradient Computation
by Grazzi, Riccardo and Franceschi, Luca and Pontil, Massimiliano and Salzo, Saverio

On the Outsized Importance of Learning Rates in Local Update Methods
by Charles, Zachary and Kone{\v c}n{\'y}, Jakub

Meta-SAC: Auto-tune the Entropy Temperature of Soft Actor-Critic via Metagradient
by Wang, Yufei and Ni, Tianwei

Meta Learning in the Continuous Time Limit
by Xu, Ruitu and Chen, Lin and Karbasi, Amin

Expert Training: Task Hardness Aware Meta-Learning for Few-Shot Classification
by Zhou, Yucan and Wang, Yu and Cai, Jianfei and Zhou, Yu and Hu, Qinghua and Wang, Weiping

MTL2L: A Context Aware Neural Optimiser
by Kuo, Nicholas I-Hsien and Harandi, Mehrtash and Fourrier, Nicolas and Walder, Christian and Ferraro, Gabriela and Suominen, Hanna

Navigating the Trade-Off between Multi-Task Learning and Learning to Multitask in Deep Neural Networks
by Ravi, Sachin and Musslick, Sebastian and Hamin, Maia and Willke, Theodore L and Cohen, Jonathan D

Balanced Meta-Softmax for Long-Tailed Visual Recognition
by Ren, Jiawei and Yu, Cunjun and Sheng, Shunan and Ma, Xiao and Zhao, Haiyu and Yi, Shuai and Li, Hongsheng

CrossTransformers: spatially-aware few-shot transfer
by Doersch, Carl and Gupta, Ankush and Zisserman, Andrew

Meta-Learning a Dynamical Language Model
by Wolf, Thomas and Chaumond, Julien and Delangue, Clement

Meta-Learning Requires Meta-Augmentation
by Rajendran, Janarthanan and Irpan, Alex and Jang, Eric

Adaptive Risk Minimization: A Meta-Learning Approach for Tackling Group Shift
by Zhang, Marvin and Marklund, Henrik and Gupta, Abhishek and Levine, Sergey and Finn, Chelsea

Meta-Learning Symmetries by Reparameterization
by Zhou, Allan and Knowles, Tom and Finn, Chelsea

Adaptive Risk Minimization: A Meta-Learning Approach for Tackling Group Shift
by Zhang, Marvin and Marklund, Henrik and Gupta, Abhishek and Levine, Sergey and Finn, Chelsea

A Brief Look at Generalization in Visual Meta-Reinforcement Learning
by Alver, Safa and Precup, Doina

Learning Representations by Stochastic Meta-Gradient Descent in Neural Networks
by Veeriah, Vivek and Zhang, Shangtong and Sutton, Richard S

PACOH: Bayes-Optimal Meta-Learning with PAC-Guarantees
by Rothfuss, Jonas and Fortuin, Vincent and Krause, Andreas

Meta-Meta-Classification for One-Shot Learning
by Chowdhury, Arkabandhu and Chaudhari, Dipak and Chaudhuri, Swarat and Jermaine, Chris

Relatedness Measures to Aid the Transfer of Building Blocks among Multiple Tasks
by Nguyen, Trung B and Browne, Will N and Zhang, Mengjie

Information-Theoretic Generalization Bounds for Meta-Learning and Applications
by Jose, Sharu Theresa and Simeone, Osvaldo

On Learning Intrinsic Rewards for Policy Gradient Methods
by Zheng, Zeyu and Oh, Junhyuk and Singh, Satinder

A Sample Complexity Separation between Non-Convex and Convex Meta-Learning
by Saunshi, Nikunj and Zhang, Yi and Khodak, Mikhail and Arora, Sanjeev

Bayesian Online Meta-Learning with Laplace Approximation
by Yap, Pau Ching and Ritter, Hippolyt and Barber, David

Meta-Reinforcement Learning for Robotic Industrial Insertion Tasks
by Schoettler, Gerrit and Nair, Ashvin and Ojea, Juan Aparicio and Levine, Sergey and Solowjow, Eugen

Continual Deep Learning by Functional Regularisation of Memorable Past
by Pan, Pingbo and Swaroop, Siddharth and Immer, Alexander and Eschenhagen, Runa and Turner, Richard E and Khan, Mohammad Emtiyaz

Jelly Bean World: A Testbed for Never-Ending Learning
by Platanios, Emmanouil Antonios and Saparov, Abulhair and Mitchell, Tom

Encouraging behavioral diversity in evolutionary robotics: an empirical study
by Mouret, J-B and Doncieux, S

Defining Benchmarks for Continual Few-Shot Learning
by Antoniou, Antreas and Patacchiola, Massimiliano and Ochal, Mateusz and Storkey, Amos

Emergent Real-World Robotic Skills via Unsupervised Off-Policy Reinforcement Learning
by Sharma, Archit and Ahn, Michael and Levine, Sergey and Kumar, Vikash and Hausman, Karol and Gu, Shixiang

Empirical Bayes Transductive Meta-Learning with Synthetic Gradients
by Hu, Shell Xu and Moreno, Pablo G and Xiao, Yang and Shen, Xi and Obozinski, Guillaume and Lawrence, Neil D and Damianou, Andreas

Evolving Inborn Knowledge For Fast Adaptation in Dynamic POMDP Problems
by Ben-Iwhiwhu, Eseoghene and Ladosz, Pawel and Dick, Jeffery and Chen, Wen-Hua and Pilly, Praveen and Soltoggio, Andrea

Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning
by Yu, Tianhe and Quillen, Deirdre and He, Zhanpeng and Julian, Ryan and Hausman, Karol and Finn, Chelsea and Levine, Sergey

Meta reinforcement learning as task inference
by Humplik, Jan and Galashov, Alexandre and Hasenclever, Leonard and Ortega, Pedro A and Teh, Yee Whye and Heess, Nicolas

Meta-Gradient Reinforcement Learning
by Xu, Zhongwen and van Hasselt, Hado and Silver, David

Self-Paced Deep Reinforcement Learning
by Klink, Pascal and D'Eramo, Carlo and Peters, Jan and Pajarinen, Joni

Scheduling the Learning Rate Via Hypergradients: New Insights and a New Algorithm
by Donini, Michele and Franceschi, Luca and Majumder, Orchid and Pontil, Massimiliano and Frasconi, Paolo

Learning Stabilizable Nonlinear Dynamics with Contraction-Based Regularization
by Singh, Sumeet and Richards, Spencer M and Sindhwani, Vikas and Slotine, Jean-Jacques E and Pavone, Marco

A Comprehensive Overview and Survey of Recent Advances in Meta-Learning
by Peng, Huimin

Learning a Formula of Interpretability to Learn Interpretable Formulas
by Virgolin, Marco and De Lorenzo, Andrea and Medvet, Eric and Randone, Francesca

Model-Based Meta-Reinforcement Learning for Flight with Suspended Payloads
by Belkhale, Suneel and Li, Rachel and Kahn, Gregory and McAllister, Rowan and Calandra, Roberto and Levine, Sergey

Frustratingly Simple Few-Shot Object Detection
by Wang, Xin and Huang, Thomas E and Darrell, Trevor and Gonzalez, Joseph E and Yu, Fisher

Meta Pseudo Labels
by Pham, Hieu and Xie, Qizhe and Dai, Zihang and Le, Quoc V

by Unknown

Finding online neural update rules by learning to remember
by Gregor, Karol

A New Meta-Baseline for Few-Shot Learning
by Chen, Yinbo and Wang, Xiaolong and Liu, Zhuang and Xu, Huijuan and Darrell, Trevor

Learning to be Global Optimizer
by Zhang, Haotian and Sun, Jianyong and Xu, Zongben

Scalable Multi-Task Imitation Learning with Autonomous Improvement
by Singh, Avi and Jang, Eric and Irpan, Alexander and Kappler, Daniel and Dalal, Murtaza and Levine, Sergey and Khansari, Mohi and Finn, Chelsea

Meta-learning for mixed linear regression
by Kong, Weihao and Somani, Raghav and Song, Zhao and Kakade, Sham and Oh, Sewoong

Provable Meta-Learning of Linear Representations
by Tripuraneni, Nilesh and Jin, Chi and Jordan, Michael I

Learning to Continually Learn
by Beaulieu, Shawn and Frati, Lapo and Miconi, Thomas and Lehman, Joel and Stanley, Kenneth O and Clune, Jeff and Cheney, Nick

PACOH: Bayes-Optimal Meta-Learning with PAC-Guarantees
by Rothfuss, Jonas and Fortuin, Vincent and Krause, Andreas

Incremental Learning for Metric-Based Meta-Learners
by Liu, Qing and Majumder, Orchid and Ravichandran, Avinash and Bhotika, Rahul and Soatto, Stefano

Hyper-Meta Reinforcement Learning with Sparse Reward
by Hua, Yun and Wang, Xiangfeng and Jin, Bo and Li, Wenhao and Yan, Junchi and He, Xiaofeng and Zha, Hongyuan

Meta-Learning across Meta-Tasks for Few-Shot Learning
by Fei, Nanyi and Lu, Zhiwu and Gao, Yizhao and Tian, Jia and Xiang, Tao and Wen, Ji-Rong

Distribution-Agnostic Model-Agnostic Meta-Learning
by Collins, Liam and Mokhtari, Aryan and Shakkottai, Sanjay

Provably Convergent Policy Gradient Methods for Model-Agnostic Meta-Reinforcement Learning
by Fallah, Alireza and Mokhtari, Aryan and Ozdaglar, Asuman

Meta-learning framework with applications to zero-shot time-series forecasting
by Oreshkin, Boris N and Carpov, Dmitri and Chapados, Nicolas and Bengio, Yoshua

A Loss-Function for Causal Machine-Learning
by Yang, I-Sheng

Self-Tuning Deep Reinforcement Learning
by Zahavy, Tom and Xu, Zhongwen and Veeriah, Vivek and Hessel, Matteo and Van Hasslet, Hado and Silver, David and Singh, Satinder

Learning Adaptive Loss for Robust Learning with Noisy Labels
by Shu, Jun and Zhao, Qian and Chen, Keyu and Xu, Zongben and Meng, Deyu

A Structured Prediction Approach for Conditional Meta-Learning
by Wang, Ruohan and Demiris, Yiannis and Ciliberto, Carlo

Curriculum in Gradient-Based Meta-Reinforcement Learning
by Mehta, Bhairav and Deleu, Tristan and Raparthy, Sharath Chandra and Pal, Chris J and Paull, Liam

Multi-Step Model-Agnostic Meta-Learning: Convergence and Improved Algorithms
by Ji, Kaiyi and Yang, Junjie and Liang, Yingbin

Local Nonparametric Meta-Learning
by Goo, Wonjoon and Niekum, Scott

Revisiting Meta-Learning as Supervised Learning
by Chao, Wei-Lun and Ye, Han-Jia and Zhan, De-Chuan and Campbell, Mark and Weinberger, Kilian Q

SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning
by Wang, Yan and Chao, Wei-Lun and Weinberger, Kilian Q and van der Maaten, Laurens

Fast and Generalized Adaptation for Few-Shot Learning
by Song, Liang and Liu, Jinlu and Qin, Yongqiang

Meta-Learning without Memorization
by Yin, Mingzhang and Tucker, George and Zhou, Mingyuan and Levine, Sergey and Finn, Chelsea

Your Classifier is Secretly an Energy Based Model and You Should Treat it Like One
by Grathwohl, Will and Wang, Kuan-Chieh and Jacobsen, J{\"o}rn-Henrik and Duvenaud, David and Norouzi, Mohammad and Swersky, Kevin

MAME : Model-Agnostic Meta-Exploration
by Gurumurthy, Swaminathan and Kumar, Sumit and Sycara, Katia

Constructing Multiple Tasks for Augmentation: Improving Neural Image Classification With K-means Features
by Gui, Tao and Qing, Lizhi and Zhang, Qi and Ye, Jiacheng and Yan, Hang and Fei, Zichu and Huang, Xuanjing

Meta Adaptation using Importance Weighted Demonstrations
by Lekkala, Kiran and Abu-El-Haija, Sami and Itti, Laurent

VIABLE: Fast Adaptation via Backpropagating Learned Loss
by Feng, Leo and Zintgraf, Luisa and Peng, Bei and Whiteson, Shimon

Decoupling Adaptation from Modeling with Meta-Optimizers for Meta Learning
by Arnold, S{\'e}bastien M R and Iqbal, Shariq and Sha, Fei

TADAM: Task dependent adaptive metric for improved few-shot learning
by Oreshkin, Boris and Rodr{\'\i}guez L{\'o}pez, Pau and Lacoste, Alexandre

Learning to Few-Shot Learn Across Diverse Natural Language Classification Tasks
by Bansal, Trapit and Jha, Rishikesh and McCallum, Andrew

Optimizing Millions of Hyperparameters by Implicit Differentiation
by Lorraine, Jonathan and Vicol, Paul and Duvenaud, David

Meta-data: Characterization of Input Features for Meta-learning
by Castiello, Ciro and Castellano, Giovanna and Fanelli, Anna Maria

Meta-Learning for Low-resource Natural Language Generation in Task-oriented Dialogue Systems
by Mi, Fei and Huang, Minlie and Zhang, Jiyong and Faltings, Boi

Domain Generalization via Model-Agnostic Learning of Semantic Features
by Dou, Qi and Castro, Daniel C and Kamnitsas, Konstantinos and Glocker, Ben

Hierarchical Expert Networks for Meta-Learning
by Hihn, Heinke and Braun, Daniel A

Online Meta-Learning on Non-convex Setting
by Zhuang, Zhenxun and Wang, Yunlong and Yu, Kezi and Lu, Songtao

Learning-to-Learn Stochastic Gradient Descent with Biased Regularization
by Denevi, Giulia and Ciliberto, Carlo and Grazzi, Riccardo and Pontil, Massimiliano

Provable Guarantees for Gradient-Based Meta-Learning
by Khodak, Mikhail and Balcan, Maria-Florina and Talwalkar, Ameet

The TCGA Meta-Dataset Clinical Benchmark
by Samiei, Mandana and W{\"u}rfl, Tobias and Deleu, Tristan and Weiss, Martin and Dutil, Francis and Fevens, Thomas and Boucher, Genevi{`e}ve and Lemieux, Sebastien and Cohen, Joseph Paul

VariBAD: A Very Good Method for Bayes-Adaptive Deep RL via Meta-Learning
by Zintgraf, Luisa and Shiarlis, Kyriacos and Igl, Maximilian and Schulze, Sebastian and Gal, Yarin and Hofmann, Katja and Whiteson, Shimon

Meta-Transfer Learning through Hard Tasks
by Sun, Qianru and Liu, Yaoyao and Chen, Zhaozheng and Chua, Tat-Seng and Schiele, Bernt

Model-Agnostic Meta-Learning using Runge-Kutta Methods
by Im, Daniel Jiwoong and Jiang, Yibo and Verma, Nakul

Improving Generalization in Meta Reinforcement Learning using Learned Objectives
by Kirsch, Louis and van Steenkiste, Sjoerd and Schmidhuber, J{\"u}rgen

Generalized Inner Loop Meta-Learning
by Grefenstette, Edward and Amos, Brandon and Yarats, Denis and Htut, Phu Mon and Molchanov, Artem and Meier, Franziska and Kiela, Douwe and Cho, Kyunghyun and Chintala, Soumith

Is Fast Adaptation All You Need?
by Javed, Khurram and Yao, Hengshuai and White, Martha

Deep Reinforcement Learning for Single-Shot Diagnosis and Adaptation in Damaged Robots
by Verma, Shresth and Nair, Haritha S and Agarwal, Gaurav and Dhar, Joydip and Shukla, Anupam

ES-MAML: Simple Hessian-Free Meta Learning
by Song, Xingyou and Gao, Wenbo and Yang, Yuxiang and Choromanski, Krzysztof and Pacchiano, Aldo and Tang, Yunhao

by Fakoor, Rasool and Chaudhari, Pratik and Soatto, Stefano and Smola, Alexander J

Efficient meta reinforcement learning via meta goal generation
by Fu, Haotian and Tang, Hongyao and Hao, Jianye

Chameleon: Learning Model Initializations Across Tasks With Different Schemas
by Brinkmeyer, Lukas and Drumond, Rafael Rego and Scholz, Randolf and Grabocka, Josif and Schmidt-Thieme, Lars

Learning Fast Adaptation with Meta Strategy Optimization
by Yu, Wenhao and Tan, Jie and Bai, Yunfei and Coumans, Erwin and Ha, Sehoon

Meta-Inverse Reinforcement Learning with Probabilistic Context Variables
by Yu, Lantao and Yu, Tianhe and Finn, Chelsea and Ermon, Stefano

Modular Meta-Learning with Shrinkage
by Chen, Yutian and Friesen, Abram L and Behbahani, Feryal and Budden, David and Hoffman, Matthew W and Doucet, Arnaud and de Freitas, Nando

Loaded DiCE: Trading off Bias and Variance in Any-Order Score Function Estimators for Reinforcement Learning
by Farquhar, Gregory and Whiteson, Shimon and Foerster, Jakob

Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
by Raghu, Aniruddh and Raghu, Maithra and Bengio, Samy and Vinyals, Oriol

by Vanschoren, Joaquin

Understanding Short-Horizon Bias in Stochastic Meta-Optimization
by Wu, Yuhuai and Ren, Mengye and Liao, Renjie and Grosse, Roger

On First-Order Meta-Learning Algorithms
by Nichol, Alex and Achiam, Joshua and Schulman, John

Towards Understanding Generalization in Gradient-Based Meta-Learning
by Guiroy, Simon and Verma, Vikas and Pal, Christopher
They empirically study the landscape of fast-adaptation in MAML. The most interesting claim is that when meta-overfitting, the loss landscape becomes flatter on test tasks.

On the Convergence Theory of Gradient-Based Model-Agnostic Meta-Learning Algorithms
by Fallah, Alireza and Mokhtari, Aryan and Ozdaglar, Asuman

Learning to Learn with Gradients
by Finn, Chelsea

Acetylcholine and memory
by Hasselmo, M E and Bower, J M

by Maudsley, Donald B

by Biggs, J B

Understanding and correcting pathologies in the training of learned optimizers
by Metz, Luke and Maheswaranathan, Niru and Nixon, Jeremy and Daniel Freeman, C and Sohl-Dickstein, Jascha
Provides many tricks (e.g. split train batch for model \& opt, average gradient estimators) for training differentiable optimizers online. They also have a couple of interesting observations specific to recurrent optimizers.

Learned Optimizers that Scale and Generalize
by Wichrowska, Olga and Maheswaranathan, Niru and Hoffman, Matthew W and Colmenarejo, Sergio Gomez and Denil, Misha and de Freitas, Nando and Sohl-Dickstein, Jascha

Using learned optimizers to make models robust to input noise
by Metz, Luke and Maheswaranathan, Niru and Shlens, Jonathon and Sohl-Dickstein, Jascha and Cubuk, Ekin D

Learning to Optimize Neural Nets
by Li, Ke and Malik, Jitendra

Meta-Learning Update Rules for Unsupervised Representation Learning
by Metz, Luke and Maheswaranathan, Niru and Cheung, Brian and Sohl-Dickstein, Jascha

Learning to Optimize
by Li, Ke and Malik, Jitendra

Learning to learn by gradient descent by gradient descent
by Andrychowicz, M and Denil, M and Gomez, S

Online Learning Rate Adaptation with Hypergradient Descent
by Baydin, Atilim Gunes and Cornish, Robert and Rubio, David Martinez and Schmidt, Mark and Wood, Frank
They adapt the learning rate of SGD by differentiating the loss of the next parameters w.r.t. the learning rate. They observe that the gradient of the learning rate is simply the inner product of the last two gradients.

Adapting Bias by Gradient Descent: An Incremental Version of Delta-Bar-Delta
by Sutton, Richard S
What's mostly interesting in this paper is the adaptation of delta-bar-delta to the online scenario. The idea of representing the learning rate as an exponential is nice. Also nice to see that the derivation suggests a full-matrix adaptive case.

Gain adaptation beats least squares
by Sutton, Richard S
This paper extends IDBD as algorithms K1 and K2, but from my quick read, it isn't clear what's the motivation for those modifications. (Seems to work in a `normalized space'', {\a} la natural gradient ?)They do work better.

Local Gain Adaptation in Stochastic Gradient Descent
by Schraudolph, Nicol N
This algorithm extends IDBD (Sutton) to the non-linear setting. Interestingly, they have a few brief discussionson the difficulties to optimize at the meta-level. (c.f. Meta-level conditioning section.) Overall, it shines light on the ground idea behind IDBD.

TIDBD: Adapting Temporal-difference Step-sizes Through Stochastic Meta-descent
by Kearney, Alex and Veeriah, Vivek and Travnik, Jaden B and Sutton, Richard S and Pilarski, Patrick M

Increased rates of convergence through learning rate adaptation
by Jacobs, Robert A
This paper argues that we need (at least) four ingredients to improve optimization of connectionist networks: 1. each parameter has its own stepsize, 2. stepsizes vary over time, 3. if consecutive gradients of a stepsize have the same sign, the stepsize should be increased, 4. conversely, if the stepsize should be decreased if its gradients have opposite signs. It also proposes to use two improvements: 1. Momentum (i.e. Polyak's heavyball), 2. delta-bar-delta (i.e. learning the stepsize). It has an interesting comment on the difficulty of learning the stepsize, and therefore comes up with a ``hack'' that outperforms momentum.

Meta-descent for Online, Continual Prediction
by Jacobsen, Andrew and Schlegel, Matthew and Linke, Cameron and Degris, Thomas and White, Adam and White, Martha
The idea is to learn the learning rate so as to minimize the norm of the gradient. They argue that for the continual learning setting, this forces the algorithm to stay ``as stable as possible''. No theorems, small-scale (but interesting) experiments.

Adaptation of learning rate parameters
by Sutton, Rich

Gradient-Based Meta-Learning with Learned Layerwise Metric and Subspace
by Lee, Yoonho and Choi, Seungjin

Meta-Learning with Warped Gradient Descent
by Flennerhag, Sebastian and Rusu, Andrei A and Pascanu, Razvan and Yin, Hujun and Hadsell, Raia

Meta-Learning via Learned Loss
by Chebotar, Yevgen and Molchanov, Artem and Bechtle, Sarah and Righetti, Ludovic and Meier, Franziska and Sukhatme, Gaurav
They learn the loss as a NN, and that loss's objective is to maximize the sum of rewards. It is provided a bunch of things, including inputs, outputs, goals.

by Park, Eunbyung and Oliva, Junier B

Alpha MAML: Adaptive Model-Agnostic Meta-Learning
by Behl, Harkirat Singh and Baydin, At{\i}l{\i}m G{\"u}ne{\c s} and Torr, Philip H S
They combine hypergradient and MAML: adapt all learning rates at all times.

Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
by Li, Zhenguo and Zhou, Fengwei and Chen, Fei and Li, Hang

ProMP: Proximal Meta-Policy Search
by Rothfuss, Jonas and Lee, Dennis and Clavera, Ignasi and Asfour, Tamim and Abbeel, Pieter

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
by Finn, Chelsea and Abbeel, Pieter and Levine, Sergey

Optimization as a model for few-shot learning
by Ravi, Sachin and Larochelle, Hugo

Fast Context Adaptation via Meta-Learning
by Zintgraf, Luisa M and Shiarlis, Kyriacos and Kurin, Vitaly and Hofmann, Katja and Whiteson, Shimon

Meta-Learning with Implicit Gradients
by Rajeswaran, Aravind and Finn, Chelsea and Kakade, Sham and Levine, Sergey

Natural Neural Networks
by Desjardins, Guillaume and Simonyan, Karen and Pascanu, Razvan and Kavukcuoglu, Koray

A Baseline for Few-Shot Image Classification
by Dhillon, Guneet S and Chaudhari, Pratik and Ravichandran, Avinash and Soatto, Stefano

by Chen, Wei-Yu and Liu, Yen-Cheng and Kira, Zsolt
Suggests that meta-learning papers haven't been tested against classical baselines. When considering those baselines, they perform better than many of the recent meta-learning techniques.

Meta-learning with differentiable closed-form solvers
by Bertinetto, Luca and Henriques, Joao F and Torr, Philip and Vedaldi, Andrea

Uncertainty in Model-Agnostic Meta-Learning using Variational Inference
by Nguyen, Cuong and Do, Thanh-Toan and Carneiro, Gustavo

Meta-Reinforcement Learning of Structured Exploration Strategies
by Gupta, Abhishek and Mendonca, Russell and Liu, Yuxuan and Abbeel, Pieter and Levine, Sergey

Metalearned Neural Memory
by Munkhdalai, Tsendsuren and Sordoni, Alessandro and Wang, Tong and Trischler, Adam

Accelerated Stochastic Approximation
by Kesten, Harry

Meta-Learning for Black-box Optimization
by Vishnu, T V and Malhotra, Pankaj and Narwariya, Jyoti and Vig, Lovekesh and Shroff, Gautam
They essentially extend the recurrent meta-learning framework in a few ways: 1. Use regret instead of objective improvement as meta-learning objective. 2. Normalize the objective so as to make it play nice with LSTMs. 3. Incorporate domain-constraints, so that the LSTM always outputs feasible solutions. All are described in page 3.

Task Agnostic Continual Learning via Meta Learning
by He, Xu and Sygnowski, Jakub and Galashov, Alexandre and Rusu, Andrei A and Teh, Yee Whye and Pascanu, Razvan

Watch, Try, Learn: Meta-Learning from Demonstrations and Reward
by Zhou, Allan and Jang, Eric and Kappler, Daniel and Herzog, Alex and Khansari, Mohi and Wohlhart, Paul and Bai, Yunfei and Kalakrishnan, Mrinal and Levine, Sergey and Finn, Chelsea

Meta-Learning Representations for Continual Learning
by Javed, Khurram and White, Martha

TapNet: Neural Network Augmented with Task-Adaptive Projection for Few-Shot Learning
by Yoon, Sung Whan and Seo, Jun and Moon, Jaekyun

Meta Reinforcement Learning with Task Embedding and Shared Policy
by Lan, Lin and Li, Zhenguo and Guan, Xiaohong and Wang, Pinghui

Hierarchically Structured Meta-learning
by Yao, Huaxiu and Wei, Ying and Huang, Junzhou and Li, Zhenhui

Curious Meta-Controller: Adaptive Alternation between Model-Based and Model-Free Control in Deep Reinforcement Learning
by Hafez, Muhammad Burhan and Weber, Cornelius and Kerzel, Matthias and Wermter, Stefan

Learning to Learn in Simulation
by Teng, Ervin and Iannucci, Bob

Meta-Learning with Differentiable Convex Optimization
by Lee, Kwonjoon and Maji, Subhransu and Ravichandran, Avinash and Soatto, Stefano

Functional Regularisation for Continual Learning
by Titsias, Michalis K and Schwarz, Jonathan and de G. Matthews, Alexander G and Pascanu, Razvan and Teh, Yee Whye

Learning to Forget for Meta-Learning
by Baik, Sungyong and Hong, Seokil and Lee, Kyoung Mu

Meta-learning of Sequential Strategies
by Ortega, Pedro A and Wang, Jane X and Rowland, Mark and Genewein, Tim and Kurth-Nelson, Zeb and Pascanu, Razvan and Heess, Nicolas and Veness, Joel and Pritzel, Alex and Sprechmann, Pablo and Jayakumar, Siddhant M and McGrath, Tom and Miller, Kevin and Azar, Mohammad and Osband, Ian and Rabinowitz, Neil and Gy{\"o}rgy, Andr{\'a}s and Chiappa, Silvia and Osindero, Simon and Teh, Yee Whye and van Hasselt, Hado and de Freitas, Nando and Botvinick, Matthew and Legg, Shane
This paper essentially provides a theoretical framework to ground the fact that recurrent meta-learning (RL^2, LLGD^2) performs Bayesian inference during adaptation.

Auto-Meta: Automated Gradient Based Meta Learner Search
by Kim, Jaehong and Lee, Sangyeul and Kim, Sungwan and Cha, Moonsu and Lee, Jung Kwon and Choi, Youngduck and Choi, Yongseok and Cho, Dong-Yeon and Kim, Jiwon

Adaptive Gradient-Based Meta-Learning Methods
by Khodak, Mikhail and Florina-Balcan, Maria and Talwalkar, Ameet

Embedded Meta-Learning: Toward more flexible deep-learning models
by Lampinen, Andrew K and McClelland, James L

Modular meta-learning
by Alet, Ferran and Lozano-P{\'e}rez, Tom{\'a}s and Kaelbling, Leslie P

MetaPred: Meta-Learning for Clinical Risk Prediction with Limited Patient Electronic Health Records
by Zhang, Xi Sheryl and Tang, Fengyi and Dodge, Hiroko and Zhou, Jiayu and Wang, Fei

Prototypical Networks for Few-shot Learning
by Snell, Jake and Swersky, Kevin and Zemel, Richard S

Meta-learners' learning dynamics are unlike learners'
by Rabinowitz, Neil C

Backpropamine: training self-modifying neural networks with differentiable neuromodulated plasticity
by Miconi, Thomas and Rawal, Aditya and Clune, Jeff and Stanley, Kenneth O

Reinforcement Learning, Fast and Slow
by Botvinick, Matthew and Ritter, Sam and Wang, Jane X and Kurth-Nelson, Zeb and Blundell, Charles and Hassabis, Demis

Been There, Done That: Meta-Learning with Episodic Recall
by Ritter, Samuel and Wang, Jane X and Kurth-Nelson, Zeb and Jayakumar, Siddhant M and Blundell, Charles and Pascanu, Razvan and Botvinick, Matthew

Guided Meta-Policy Search
by Mendonca, Russell and Gupta, Abhishek and Kralev, Rosen and Abbeel, Pieter and Levine, Sergey and Finn, Chelsea

Hierarchical Meta Learning
by Zou, Yingtian and Feng, Jiashi

A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms
by Bengio, Yoshua and Deleu, Tristan and Rahaman, Nasim and Ke, Rosemary and Lachapelle, S{\'e}bastien and Bilaniuk, Olexa and Goyal, Anirudh and Pal, Christopher

Generalize Across Tasks: Efficient Algorithms for Linear Representation Learning
by Bullins, Brian and Hazan, Elad and Kalai, Adam and Livni, Roi

Incremental Learning-to-Learn with Statistical Guarantees
by Denevi, Giulia and Ciliberto, Carlo and Stamos, Dimitris and Pontil, Massimiliano

A Model of Inductive Bias Learning
by Baxter, J

Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic Context Variables
by Rakelly, Kate and Zhou, Aurick and Quillen, Deirdre and Finn, Chelsea and Levine, Sergey

Continual Learning with Tiny Episodic Memories
by Chaudhry, Arslan and Rohrbach, Marcus and Elhoseiny, Mohamed and Ajanthan, Thalaiyasingam and Dokania, Puneet K and Torr, Philip H S and Ranzato, Marc'aurelio

Online Meta-Learning
by Finn, Chelsea and Rajeswaran, Aravind and Kakade, Sham and Levine, Sergey

Modulating transfer between tasks in gradient-based meta-learning
by Grant, Erin and Jerfel, Ghassen and Heller, Katherine and Griffiths, Thomas L

Learning to Adapt in Dynamic, Real-World Environments Through Meta-Reinforcement Learning
by Nagabandi, Anusha and Clavera, Ignasi and Liu, Simin and Fearing, Ronald S and Abbeel, Pieter and Levine, Sergey and Finn, Chelsea

Meta-Learning with Latent Embedding Optimization
by Rusu, Andrei A and Rao, Dushyant and Sygnowski, Jakub and Vinyals, Oriol and Pascanu, Razvan and Osindero, Simon and Hadsell, Raia

Learning to Generalize: Meta-Learning for Domain Generalization
by Li, Da and Yang, Yongxin and Song, Yi-Zhe and Hospedales, Timothy M

Some Considerations on Learning to Explore via Meta-Reinforcement Learning
by Stadie, Bradly C and Yang, Ge and Houthooft, Rein and Chen, Xi and Duan, Yan and Wu, Yuhuai and Abbeel, Pieter and Sutskever, Ilya

How to train your MAML
by Antoniou, Antreas and Edwards, Harrison and Storkey, Amos

Bayesian Model-Agnostic Meta-Learning
by Kim, Taesup and Yoon, Jaesik and Dia, Ousmane and Kim, Sungwoong and Bengio, Yoshua and Ahn, Sungjin

Probabilistic Model-Agnostic Meta-Learning
by Finn, Chelsea and Xu, Kelvin and Levine, Sergey

The effects of negative adaptation in Model-Agnostic Meta-Learning
by Deleu, Tristan and Bengio, Yoshua

Memory-based Parameter Adaptation
by Sprechmann, Pablo and Jayakumar, Siddhant M and Rae, Jack W and Pritzel, Alexander and Badia, Adri{`a} Puigdom{`e}nech and Uria, Benigno and Vinyals, Oriol and Hassabis, Demis and Pascanu, Razvan and Blundell, Charles

Deep Meta-Learning: Learning to Learn in the Concept Space
by Zhou, Fengwei and Wu, Bin and Li, Zhenguo

Deep Prior
by Lacoste, Alexandre and Boquet, Thomas and Rostamzadeh, Negar and Oreshkin, Boris and Chung, Wonchang and Krueger, David

Recasting Gradient-Based Meta-Learning as Hierarchical Bayes
by Grant, Erin and Finn, Chelsea and Levine, Sergey and Darrell, Trevor and Griffiths, Thomas

WNGrad: Learn the Learning Rate in Gradient Descent
by Wu, Xiaoxia and Ward, Rachel and Bottou, L{\'e}on

Learning to Learn
by Finn, Chelsea

Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments
by Al-Shedivat, Maruan and Bansal, Trapit and Burda, Yuri and Sutskever, Ilya and Mordatch, Igor and Abbeel, Pieter

Submission Form