Search Results for author: Razvan Pascanu

Found 81 papers, 36 papers with code

Architecture Matters in Continual Learning

no code implementations1 Feb 2022 Seyed Iman Mirzadeh, Arslan Chaudhry, Dong Yin, Timothy Nguyen, Razvan Pascanu, Dilan Gorur, Mehrdad Farajtabar

However, in this work, we show that the choice of architecture can significantly impact the continual learning performance, and different architectures lead to different trade-offs between the ability to remember previous tasks and learning new ones.

Continual Learning

Pushing the limits of self-supervised ResNets: Can we outperform supervised learning without labels on ImageNet?

no code implementations13 Jan 2022 Nenad Tomasev, Ioana Bica, Brian McWilliams, Lars Buesing, Razvan Pascanu, Charles Blundell, Jovana Mitrovic

Despite recent progress made by self-supervised methods in representation learning with residual networks, they still underperform supervised learning on the ImageNet classification benchmark, limiting their applicability in performance-critical settings.

Representation Learning Self-Supervised Image Classification +2

Wide Neural Networks Forget Less Catastrophically

no code implementations21 Oct 2021 Seyed Iman Mirzadeh, Arslan Chaudhry, Dong Yin, Huiyi Hu, Razvan Pascanu, Dilan Gorur, Mehrdad Farajtabar

A primary focus area in continual learning research is alleviating the "catastrophic forgetting" problem in neural networks by designing new algorithms that are more robust to the distribution shifts.

Continual Learning

Powerpropagation: A sparsity inducing weight reparameterisation

2 code implementations NeurIPS 2021 Jonathan Schwarz, Siddhant M. Jayakumar, Razvan Pascanu, Peter E. Latham, Yee Whye Teh

The training of sparse neural networks is becoming an increasingly important tool for reducing the computational footprint of models at training and evaluation, as well enabling the effective scaling up of models.

On the Role of Optimization in Double Descent: A Least Squares Study

no code implementations NeurIPS 2021 Ilja Kuzborskij, Csaba Szepesvári, Omar Rivasplata, Amal Rannen-Triki, Razvan Pascanu

Empirically it has been observed that the performance of deep neural networks steadily improves as we increase model size, contradicting the classical view on overfitting and generalization.

Task-agnostic Continual Learning with Hybrid Probabilistic Models

no code implementations ICML Workshop INNF 2021 Polina Kirichenko, Mehrdad Farajtabar, Dushyant Rao, Balaji Lakshminarayanan, Nir Levine, Ang Li, Huiyi Hu, Andrew Gordon Wilson, Razvan Pascanu

Learning new tasks continuously without forgetting on a constantly changing data distribution is essential for real-world problems but extremely challenging for modern deep learning.

Anomaly Detection Continual Learning +1

Test Sample Accuracy Scales with Training Sample Density in Neural Networks

1 code implementation15 Jun 2021 Xu Ji, Razvan Pascanu, Devon Hjelm, Balaji Lakshminarayanan, Andrea Vedaldi

Intuitively, one would expect the accuracy of a trained neural network's prediction on a test sample to correlate with how densely that sample is surrounded by seen training samples in representation space.

Image Classification

Top-KAST: Top-K Always Sparse Training

1 code implementation NeurIPS 2020 Siddhant M. Jayakumar, Razvan Pascanu, Jack W. Rae, Simon Osindero, Erich Elsen

Sparse neural networks are becoming increasingly important as the field seeks to improve the performance of existing models by scaling them up, while simultaneously trying to reduce power consumption and computational footprint.

Language Modelling

A study on the plasticity of neural networks

no code implementations31 May 2021 Tudor Berariu, Wojciech Czarnecki, Soham De, Jorg Bornschein, Samuel Smith, Razvan Pascanu, Claudia Clopath

One aim shared by multiple settings, such as continual learning or transfer learning, is to leverage previously acquired knowledge to converge faster on the current task.

Continual Learning Transfer Learning

Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error

no code implementations27 May 2021 Stanislav Fort, Andrew Brock, Razvan Pascanu, Soham De, Samuel L. Smith

In this work, we provide a detailed empirical evaluation of how the number of augmentation samples per unique image influences model performance on held out data when training deep ResNets.

Data Augmentation Image Classification

Continual World: A Robotic Benchmark For Continual Reinforcement Learning

1 code implementation NeurIPS 2021 Maciej Wołczyk, Michał Zając, Razvan Pascanu, Łukasz Kuciński, Piotr Miłoś

Continual learning (CL) -- the ability to continuously learn, building on previously acquired knowledge -- is a natural requirement for long-lived autonomous reinforcement learning (RL) agents.

Continual Learning reinforcement-learning

Spectral Normalisation for Deep Reinforcement Learning: an Optimisation Perspective

1 code implementation11 May 2021 Florin Gogianu, Tudor Berariu, Mihaela Rosca, Claudia Clopath, Lucian Busoniu, Razvan Pascanu

We conduct ablation studies to disentangle the various effects normalisation has on the learning dynamics and show that is sufficient to modulate the parameter updates to recover most of the performance of spectral normalisation.


Regularized Behavior Value Estimation

no code implementations17 Mar 2021 Caglar Gulcehre, Sergio Gómez Colmenarejo, Ziyu Wang, Jakub Sygnowski, Thomas Paine, Konrad Zolna, Yutian Chen, Matthew Hoffman, Razvan Pascanu, Nando de Freitas

Due to bootstrapping, these errors get amplified during training and can lead to divergence, thereby crippling learning.

Offline RL

Addressing Extrapolation Error in Deep Offline Reinforcement Learning

no code implementations1 Jan 2021 Caglar Gulcehre, Sergio Gómez Colmenarejo, Ziyu Wang, Jakub Sygnowski, Thomas Paine, Konrad Zolna, Yutian Chen, Matthew Hoffman, Razvan Pascanu, Nando de Freitas

These errors can be compounded by bootstrapping when the function approximator overestimates, leading the value function to *grow unbounded*, thereby crippling learning.

Offline RL reinforcement-learning

Behavior Priors for Efficient Reinforcement Learning

no code implementations27 Oct 2020 Dhruva Tirumala, Alexandre Galashov, Hyeonwoo Noh, Leonard Hasenclever, Razvan Pascanu, Jonathan Schwarz, Guillaume Desjardins, Wojciech Marian Czarnecki, Arun Ahuja, Yee Whye Teh, Nicolas Heess

In this work we consider how information and architectural constraints can be combined with ideas from the probabilistic modeling literature to learn behavior priors that capture the common movement and interaction patterns that are shared across a set of related tasks or contexts.

Continuous Control Hierarchical Reinforcement Learning +2

Linear Mode Connectivity in Multitask and Continual Learning

1 code implementation ICLR 2021 Seyed Iman Mirzadeh, Mehrdad Farajtabar, Dilan Gorur, Razvan Pascanu, Hassan Ghasemzadeh

Continual (sequential) training and multitask (simultaneous) training are often attempting to solve the same overall objective: to find a solution that performs well on all considered tasks.

Continual Learning

Temporal Difference Uncertainties as a Signal for Exploration

no code implementations5 Oct 2020 Sebastian Flennerhag, Jane X. Wang, Pablo Sprechmann, Francesco Visin, Alexandre Galashov, Steven Kapturowski, Diana L. Borsa, Nicolas Heess, Andre Barreto, Razvan Pascanu

Instead, we incorporate it as an intrinsic reward and treat exploration as a separate learning problem, induced by the agent's temporal difference uncertainties.

Understanding the Role of Training Regimes in Continual Learning

4 code implementations NeurIPS 2020 Seyed Iman Mirzadeh, Mehrdad Farajtabar, Razvan Pascanu, Hassan Ghasemzadeh

However, there has been limited prior work extensively analyzing the impact that different training regimes -- learning rate, batch size, regularization method-- can have on forgetting.

Continual Learning

Pointer Graph Networks

no code implementations NeurIPS 2020 Petar Veličković, Lars Buesing, Matthew C. Overlan, Razvan Pascanu, Oriol Vinyals, Charles Blundell

This static input structure is often informed purely by insight of the machine learning practitioner, and might not be optimal for the actual task the GNN is solving.

Multiplicative Interactions and Where to Find Them

no code implementations ICLR 2020 Siddhant M. Jayakumar, Wojciech M. Czarnecki, Jacob Menick, Jonathan Schwarz, Jack Rae, Simon Osindero, Yee Whye Teh, Tim Harley, Razvan Pascanu

We explore the role of multiplicative interaction as a unifying framework to describe a range of classical and modern neural network architectural motifs, such as gating, attention layers, hypernetworks, and dynamic convolutions amongst others.

A Deep Neural Network's Loss Surface Contains Every Low-dimensional Pattern

no code implementations16 Dec 2019 Wojciech Marian Czarnecki, Simon Osindero, Razvan Pascanu, Max Jaderberg

The work "Loss Landscape Sightseeing with Multi-Point Optimization" (Skorokhodov and Burtsev, 2019) demonstrated that one can empirically find arbitrary 2D binary patterns inside loss surfaces of popular neural networks.

Continual Unsupervised Representation Learning

1 code implementation NeurIPS 2019 Dushyant Rao, Francesco Visin, Andrei A. Rusu, Yee Whye Teh, Razvan Pascanu, Raia Hadsell

Continual learning aims to improve the ability of modern learning systems to deal with non-stationary distributions, typically by attempting to learn a series of tasks sequentially.

Continual Learning Representation Learning

Stabilizing Transformers for Reinforcement Learning

5 code implementations ICML 2020 Emilio Parisotto, H. Francis Song, Jack W. Rae, Razvan Pascanu, Caglar Gulcehre, Siddhant M. Jayakumar, Max Jaderberg, Raphael Lopez Kaufman, Aidan Clark, Seb Noury, Matthew M. Botvinick, Nicolas Heess, Raia Hadsell

Harnessing the transformer's ability to process long time horizons of information could provide a similar performance boost in partially observable reinforcement learning (RL) domains, but the large-scale transformers used in NLP have yet to be successfully applied to the RL setting.

General Reinforcement Learning Language Modelling +2

Meta-Learning with Warped Gradient Descent

1 code implementation ICLR 2020 Sebastian Flennerhag, Andrei A. Rusu, Razvan Pascanu, Francesco Visin, Hujun Yin, Raia Hadsell

On the other hand, approaches that try to control a gradient-based update rule typically resort to computing gradients through the learning process to obtain their meta-gradients, leading to methods that can not scale beyond few-shot task adaptation.

Few-Shot Learning

Task Agnostic Continual Learning via Meta Learning

no code implementations ICML Workshop LifelongML 2020 Xu He, Jakub Sygnowski, Alexandre Galashov, Andrei A. Rusu, Yee Whye Teh, Razvan Pascanu

One particular formalism that studies learning under non-stationary distribution is provided by continual learning, where the non-stationarity is imposed by a sequence of distinct tasks.

Continual Learning Meta-Learning

Deep reinforcement learning with relational inductive biases

no code implementations ICLR 2019 Vinicius Zambaldi, David Raposo, Adam Santoro, Victor Bapst, Yujia Li, Igor Babuschkin, Karl Tuyls, David Reichert, Timothy Lillicrap, Edward Lockhart, Murray Shanahan, Victoria Langston, Razvan Pascanu, Matthew Botvinick, Oriol Vinyals, Peter Battaglia

We introduce an approach for augmenting model-free deep reinforcement learning agents with a mechanism for relational reasoning over structured representations, which improves performance, learning efficiency, generalization, and interpretability.

reinforcement-learning Relational Reasoning +2

Ray Interference: a Source of Plateaus in Deep Reinforcement Learning

no code implementations25 Apr 2019 Tom Schaul, Diana Borsa, Joseph Modayil, Razvan Pascanu

Rather than proposing a new method, this paper investigates an issue present in existing learning algorithms.


Exploiting Hierarchy for Learning and Transfer in KL-regularized RL

no code implementations18 Mar 2019 Dhruva Tirumala, Hyeonwoo Noh, Alexandre Galashov, Leonard Hasenclever, Arun Ahuja, Greg Wayne, Razvan Pascanu, Yee Whye Teh, Nicolas Heess

As reinforcement learning agents are tasked with solving more challenging and diverse tasks, the ability to incorporate prior knowledge into the learning system and to exploit reusable structure in solution space is likely to become increasingly important.

Continuous Control reinforcement-learning

Distilling Policy Distillation

no code implementations6 Feb 2019 Wojciech Marian Czarnecki, Razvan Pascanu, Simon Osindero, Siddhant M. Jayakumar, Grzegorz Swirszcz, Max Jaderberg

The transfer of knowledge from one policy to another is an important tool in Deep Reinforcement Learning.


Functional Regularisation for Continual Learning with Gaussian Processes

1 code implementation ICLR 2020 Michalis K. Titsias, Jonathan Schwarz, Alexander G. de G. Matthews, Razvan Pascanu, Yee Whye Teh

We introduce a framework for Continual Learning (CL) based on Bayesian inference over the function space rather than the parameters of a deep neural network.

Bayesian Inference Continual Learning +2

Adapting Auxiliary Losses Using Gradient Similarity

1 code implementation5 Dec 2018 Yunshu Du, Wojciech M. Czarnecki, Siddhant M. Jayakumar, Mehrdad Farajtabar, Razvan Pascanu, Balaji Lakshminarayanan

One approach to deal with the statistical inefficiency of neural networks is to rely on auxiliary losses that help to build useful representations.

Atari Games reinforcement-learning

Meta-Learning with Latent Embedding Optimization

4 code implementations ICLR 2019 Andrei A. Rusu, Dushyant Rao, Jakub Sygnowski, Oriol Vinyals, Razvan Pascanu, Simon Osindero, Raia Hadsell

We show that it is possible to bypass these limitations by learning a data-dependent latent generative representation of model parameters, and performing gradient-based meta-learning in this low-dimensional latent space.

Few-Shot Learning

Mix & Match - Agent Curricula for Reinforcement Learning

no code implementations ICML 2018 Wojciech Czarnecki, Siddhant Jayakumar, Max Jaderberg, Leonard Hasenclever, Yee Whye Teh, Nicolas Heess, Simon Osindero, Razvan Pascanu

We introduce Mix and match (M&M) – a training framework designed to facilitate rapid and effective learning in RL agents that would be too slow or too challenging to train otherwise. The key innovation is a procedure that allows us to automatically form a curriculum over agents.


Relational Deep Reinforcement Learning

7 code implementations5 Jun 2018 Vinicius Zambaldi, David Raposo, Adam Santoro, Victor Bapst, Yujia Li, Igor Babuschkin, Karl Tuyls, David Reichert, Timothy Lillicrap, Edward Lockhart, Murray Shanahan, Victoria Langston, Razvan Pascanu, Matthew Botvinick, Oriol Vinyals, Peter Battaglia

We introduce an approach for deep reinforcement learning (RL) that improves upon the efficiency, generalization capacity, and interpretability of conventional approaches through structured perception and relational reasoning.

reinforcement-learning Relational Reasoning +2

Mix&Match - Agent Curricula for Reinforcement Learning

no code implementations5 Jun 2018 Wojciech Marian Czarnecki, Siddhant M. Jayakumar, Max Jaderberg, Leonard Hasenclever, Yee Whye Teh, Simon Osindero, Nicolas Heess, Razvan Pascanu

(2) We further show that M&M can be used successfully to progress through a curriculum of architectural variants defining an agents internal state.


Hyperbolic Attention Networks

no code implementations ICLR 2019 Caglar Gulcehre, Misha Denil, Mateusz Malinowski, Ali Razavi, Razvan Pascanu, Karl Moritz Hermann, Peter Battaglia, Victor Bapst, David Raposo, Adam Santoro, Nando de Freitas

We introduce hyperbolic attention networks to endow neural networks with enough capacity to match the complexity of data with hierarchical and power-law structure.

Machine Translation Question Answering +2

Been There, Done That: Meta-Learning with Episodic Recall

1 code implementation ICML 2018 Samuel Ritter, Jane. X. Wang, Zeb Kurth-Nelson, Siddhant M. Jayakumar, Charles Blundell, Razvan Pascanu, Matthew Botvinick

Meta-learning agents excel at rapidly learning new tasks from open-ended task distributions; yet, they forget what they learn about each task as soon as the next begins.


Progress & Compress: A scalable framework for continual learning

no code implementations ICML 2018 Jonathan Schwarz, Jelena Luketina, Wojciech M. Czarnecki, Agnieszka Grabska-Barwinska, Yee Whye Teh, Razvan Pascanu, Raia Hadsell

This is achieved by training a network with two components: A knowledge base, capable of solving previously encountered problems, which is connected to an active column that is employed to efficiently learn the current task.

Active Learning Atari Games +1

Low-pass Recurrent Neural Networks - A memory architecture for longer-term correlation discovery

no code implementations13 May 2018 Thomas Stepleton, Razvan Pascanu, Will Dabney, Siddhant M. Jayakumar, Hubert Soyer, Remi Munos

Reinforcement learning (RL) agents performing complex tasks must be able to remember observations and actions across sizable time intervals.

Block Mean Approximation for Efficient Second Order Optimization

no code implementations16 Apr 2018 Yao Lu, Mehrtash Harandi, Richard Hartley, Razvan Pascanu

Advanced optimization algorithms such as Newton method and AdaGrad benefit from second order derivative or second order statistics to achieve better descent directions and faster convergence rates.

Learning Deep Generative Models of Graphs

no code implementations ICLR 2018 Yujia Li, Oriol Vinyals, Chris Dyer, Razvan Pascanu, Peter Battaglia

Graphs are fundamental data structures which concisely capture the relational structure in many important real-world domains, such as knowledge graphs, physical and social interactions, language, and chemistry.

Graph Generation Knowledge Graphs

Model compression via distillation and quantization

5 code implementations ICLR 2018 Antonio Polino, Razvan Pascanu, Dan Alistarh

Deep neural networks (DNNs) continue to make significant advances, solving tasks from image classification to translation or reinforcement learning.

Model Compression Quantization

Visual Interaction Networks: Learning a Physics Simulator from Video

no code implementations NeurIPS 2017 Nicholas Watters, Daniel Zoran, Theophane Weber, Peter Battaglia, Razvan Pascanu, Andrea Tacchetti

We introduce the Visual Interaction Network, a general-purpose model for learning the dynamics of a physical system from raw visual observations.

Decision Making

Learning model-based planning from scratch

1 code implementation19 Jul 2017 Razvan Pascanu, Yujia Li, Oriol Vinyals, Nicolas Heess, Lars Buesing, Sebastien Racanière, David Reichert, Théophane Weber, Daan Wierstra, Peter Battaglia

Here we introduce the "Imagination-based Planner", the first model-based, sequential decision-making agent that can learn to construct, evaluate, and execute plans.

Continuous Control Decision Making

Sobolev Training for Neural Networks

no code implementations NeurIPS 2017 Wojciech Marian Czarnecki, Simon Osindero, Max Jaderberg, Grzegorz Świrszcz, Razvan Pascanu

In many cases we only have access to input-output pairs from the ground truth, however it is becoming more common to have access to derivatives of the target output with respect to the input - for example when the ground truth function is itself a neural network such as in network compression or distillation.

Visual Interaction Networks

2 code implementations5 Jun 2017 Nicholas Watters, Andrea Tacchetti, Theophane Weber, Razvan Pascanu, Peter Battaglia, Daniel Zoran

We found that from just six input video frames the Visual Interaction Network can generate accurate future trajectories of hundreds of time steps on a wide range of physical systems.

Decision Making

Metacontrol for Adaptive Imagination-Based Optimization

1 code implementation7 May 2017 Jessica B. Hamrick, Andrew J. Ballard, Razvan Pascanu, Oriol Vinyals, Nicolas Heess, Peter W. Battaglia

The metacontroller component is a model-free reinforcement learning agent, which decides both how many iterations of the optimization procedure to run, as well as which model to consult on each iteration.

Decision Making

Sharp Minima Can Generalize For Deep Nets

no code implementations ICML 2017 Laurent Dinh, Razvan Pascanu, Samy Bengio, Yoshua Bengio

Despite their overwhelming capacity to overfit, deep learning architectures tend to generalize relatively well to unseen data, allowing them to be deployed in practice.

Interaction Networks for Learning about Objects, Relations and Physics

6 code implementations NeurIPS 2016 Peter W. Battaglia, Razvan Pascanu, Matthew Lai, Danilo Rezende, Koray Kavukcuoglu

Here we introduce the interaction network, a model which can reason about how objects in complex systems interact, supporting dynamical predictions, as well as inferences about the abstract properties of the system.

Local minima in training of neural networks

1 code implementation19 Nov 2016 Grzegorz Swirszcz, Wojciech Marian Czarnecki, Razvan Pascanu

Given that deep networks are highly nonlinear systems optimized by local gradient methods, why do they not seem to be affected by bad local minima?

Sim-to-Real Robot Learning from Pixels with Progressive Nets

no code implementations13 Oct 2016 Andrei A. Rusu, Mel Vecerik, Thomas Rothörl, Nicolas Heess, Razvan Pascanu, Raia Hadsell

The progressive net approach is a general framework that enables reuse of everything from low-level visual features to high-level policies for transfer to new tasks, enabling a compositional, yet simple, approach to building complex skills.


Progressive Neural Networks

9 code implementations15 Jun 2016 Andrei A. Rusu, Neil C. Rabinowitz, Guillaume Desjardins, Hubert Soyer, James Kirkpatrick, Koray Kavukcuoglu, Razvan Pascanu, Raia Hadsell

Learning to solve complex sequences of tasks--while both leveraging transfer and avoiding catastrophic forgetting--remains a key obstacle to achieving human-level intelligence.

Continual Learning reinforcement-learning

Theano: A Python framework for fast computation of mathematical expressions

1 code implementation9 May 2016 The Theano Development Team, Rami Al-Rfou, Guillaume Alain, Amjad Almahairi, Christof Angermueller, Dzmitry Bahdanau, Nicolas Ballas, Frédéric Bastien, Justin Bayer, Anatoly Belikov, Alexander Belopolsky, Yoshua Bengio, Arnaud Bergeron, James Bergstra, Valentin Bisson, Josh Bleecher Snyder, Nicolas Bouchard, Nicolas Boulanger-Lewandowski, Xavier Bouthillier, Alexandre de Brébisson, Olivier Breuleux, Pierre-Luc Carrier, Kyunghyun Cho, Jan Chorowski, Paul Christiano, Tim Cooijmans, Marc-Alexandre Côté, Myriam Côté, Aaron Courville, Yann N. Dauphin, Olivier Delalleau, Julien Demouth, Guillaume Desjardins, Sander Dieleman, Laurent Dinh, Mélanie Ducoffe, Vincent Dumoulin, Samira Ebrahimi Kahou, Dumitru Erhan, Ziye Fan, Orhan Firat, Mathieu Germain, Xavier Glorot, Ian Goodfellow, Matt Graham, Caglar Gulcehre, Philippe Hamel, Iban Harlouchet, Jean-Philippe Heng, Balázs Hidasi, Sina Honari, Arjun Jain, Sébastien Jean, Kai Jia, Mikhail Korobov, Vivek Kulkarni, Alex Lamb, Pascal Lamblin, Eric Larsen, César Laurent, Sean Lee, Simon Lefrancois, Simon Lemieux, Nicholas Léonard, Zhouhan Lin, Jesse A. Livezey, Cory Lorenz, Jeremiah Lowin, Qianli Ma, Pierre-Antoine Manzagol, Olivier Mastropietro, Robert T. McGibbon, Roland Memisevic, Bart van Merriënboer, Vincent Michalski, Mehdi Mirza, Alberto Orlandi, Christopher Pal, Razvan Pascanu, Mohammad Pezeshki, Colin Raffel, Daniel Renshaw, Matthew Rocklin, Adriana Romero, Markus Roth, Peter Sadowski, John Salvatier, François Savard, Jan Schlüter, John Schulman, Gabriel Schwartz, Iulian Vlad Serban, Dmitriy Serdyuk, Samira Shabanian, Étienne Simon, Sigurd Spieckermann, S. Ramana Subramanyam, Jakub Sygnowski, Jérémie Tanguay, Gijs van Tulder, Joseph Turian, Sebastian Urban, Pascal Vincent, Francesco Visin, Harm de Vries, David Warde-Farley, Dustin J. Webb, Matthew Willson, Kelvin Xu, Lijun Xue, Li Yao, Saizheng Zhang, Ying Zhang

Since its introduction, it has been one of the most used CPU and GPU mathematical compilers - especially in the machine learning community - and has shown steady performance improvements.

Dimensionality Reduction General Classification

Policy Distillation

1 code implementation19 Nov 2015 Andrei A. Rusu, Sergio Gomez Colmenarejo, Caglar Gulcehre, Guillaume Desjardins, James Kirkpatrick, Razvan Pascanu, Volodymyr Mnih, Koray Kavukcuoglu, Raia Hadsell

Policies for complex visual tasks have been successfully learned with deep reinforcement learning, using an approach called deep Q-networks (DQN), but relatively large (task-specific) networks and extensive training are needed to achieve good performance.


Natural Neural Networks

1 code implementation NeurIPS 2015 Guillaume Desjardins, Karen Simonyan, Razvan Pascanu, Koray Kavukcuoglu

We introduce Natural Neural Networks, a novel family of algorithms that speed up convergence by adapting their internal representation during training to improve conditioning of the Fisher matrix.

online learning

Identifying and attacking the saddle point problem in high-dimensional non-convex optimization

3 code implementations NeurIPS 2014 Yann Dauphin, Razvan Pascanu, Caglar Gulcehre, Kyunghyun Cho, Surya Ganguli, Yoshua Bengio

Gradient descent or quasi-Newton methods are almost ubiquitously used to perform such minimizations, and it is often thought that a main source of difficulty for these local methods to find the global minimum is the proliferation of local minima with much higher error than the global minimum.

On the saddle point problem for non-convex optimization

no code implementations19 May 2014 Razvan Pascanu, Yann N. Dauphin, Surya Ganguli, Yoshua Bengio

Gradient descent or quasi-Newton methods are almost ubiquitously used to perform such minimizations, and it is often thought that a main source of difficulty for the ability of these local methods to find the global minimum is the proliferation of local minima with much higher error than the global minimum.

On the Number of Linear Regions of Deep Neural Networks

no code implementations NeurIPS 2014 Guido Montúfar, Razvan Pascanu, Kyunghyun Cho, Yoshua Bengio

We study the complexity of functions computable by deep feedforward neural networks with piecewise linear activations in terms of the symmetries and the number of linear regions that they have.

How to Construct Deep Recurrent Neural Networks

no code implementations20 Dec 2013 Razvan Pascanu, Caglar Gulcehre, Kyunghyun Cho, Yoshua Bengio

Based on this observation, we propose two novel architectures of a deep RNN which are orthogonal to an earlier attempt of stacking multiple recurrent layers to build a deep RNN (Schmidhuber, 1992; El Hihi and Bengio, 1996).

Language Modelling

On the number of response regions of deep feed forward networks with piece-wise linear activations

no code implementations20 Dec 2013 Razvan Pascanu, Guido Montufar, Yoshua Bengio

For a $k$ layer model with $n$ hidden units on each layer it is $\Omega(\left\lfloor {n}/{n_0}\right\rfloor^{k-1}n^{n_0})$.

Learned-Norm Pooling for Deep Feedforward and Recurrent Neural Networks

no code implementations7 Nov 2013 Caglar Gulcehre, Kyunghyun Cho, Razvan Pascanu, Yoshua Bengio

In this paper we propose and investigate a novel nonlinear unit, called $L_p$ unit, for deep neural networks.

Object Recognition

Revisiting Natural Gradient for Deep Networks

no code implementations16 Jan 2013 Razvan Pascanu, Yoshua Bengio

We evaluate natural gradient, an algorithm originally proposed in Amari (1997), for learning deep models.

Advances in Optimizing Recurrent Networks

no code implementations4 Dec 2012 Yoshua Bengio, Nicolas Boulanger-Lewandowski, Razvan Pascanu

After a more than decade-long period of relatively little research activity in the area of recurrent neural networks, several new developments will be reviewed here that have allowed substantial progress both in understanding and in technical solutions towards more efficient training of recurrent networks.

Theano: new features and speed improvements

no code implementations23 Nov 2012 Frédéric Bastien, Pascal Lamblin, Razvan Pascanu, James Bergstra, Ian Goodfellow, Arnaud Bergeron, Nicolas Bouchard, David Warde-Farley, Yoshua Bengio

Theano is a linear algebra compiler that optimizes a user's symbolically-specified mathematical computations to produce efficient low-level implementations.

On the difficulty of training Recurrent Neural Networks

2 code implementations21 Nov 2012 Razvan Pascanu, Tomas Mikolov, Yoshua Bengio

There are two widely known issues with properly training Recurrent Neural Networks, the vanishing and the exploding gradient problems detailed in Bengio et al. (1994).

Cannot find the paper you are looking for? You can Submit a new open access paper.