Active Long Term Memory Networks

By Tommaso Furlanello, Jiaping Zhao, Andrew M. Saxe, Laurent Itti, Bosco S. Tjan. 2016

44 citations recorded by the source corpus.

Continual Learning in artificial neural networks suffers from interference and forgetting when different tasks are learned sequentially. This paper introduces the Active Long Term Memory Networks (A-LTM), a model of sequential multi-task deep learning that is able to maintain previously learned association between sensory input and behavioral output while acquiring knew knowledge. A-LTM exploits the non-convex nature of deep neural networks and actively maintains knowledge of previously learned, inactive tasks using a distillation loss. Distortions of the learned input-output map are penalized but hidden layers are free to transverse towards new local optima that are more favorable for the multi-task objective. We re-frame the McClelland's seminal Hippocampal theory with respect to Catastrophic Inference (CI) behavior exhibited by modern deep architectures trained with back-propagation and inhomogeneous sampling of latent factors across epochs. We present empirical results of non-trivial CI during continual learning in Deep Linear Networks trained on the same task, in Convolutional Neural Networks when the task shifts from predicting semantic to graphical factors and during domain adaptation from simple to complex environments. We present results of the A-LTM model's ability to maintain viewpoint recognition learned in the highly controlled iLab-20M dataset with 10 object categories and 88 camera viewpoints, while adapting to the unstructured domain of Imagenet with 1,000 object categories.

Subfields: Artificial Intelligence, Electrical and Electronic Engineering, Model Architectures

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Electrical and Electronic Engineering

In Model Architectures

Citations in the corpus

Growing a Brain: Fine-Tuning by Increasing Model Capacity (2017, background): “Past work has explored strategies for preserving accuracy on the source task [22, 8], while our primary focus is on improving accuracy on the target task.”
Born Again Neural Networks (2018, background): “A number of recent papers (Furlanello et al., 2016; Li & Hoiem, 2016; Shin et al., 2017) employ KD for the purpose of minimizing forgetting in continual learning.”
Learning without Forgetting (2016, methodology): “The experiments in A-LTM [8] find much larger loss due to fine-tuning than we do, and the paper concludes that maintaining the data from the original task is necessary”
Knowledge Flow: Improve Upon Your Teachers (2019, methodology): “Similar techniques have been developed by Furlanello et al. (2016); Jung et al. (2016).”
CORe50: a New Dataset and Benchmark for Continuous Object Recognition (2017, methodology): “…approaches have been proposed such as: Learning without Forgetting [13], Progressive Neural Networks [23], Active Long Term Memory Networks [7], Adaptive Convolutional Neural Network [29], PathNet [5], Incremental Regularized Least Squares [3], Elastic Weight Consolidation [11],…”
Policy Consolidation for Continual Reinforcement Learning (2019, background): “, 2018), knowledge is distilled unidirectionally from a flexible network to a more stable one, and vice versa in (Furlanello et al., 2016).”
Closed-Loop GAN for continual Learning (2018, methodology): “To guarantee optimal performance on sequential tasks, the conventional solution has been to store all learned (old) data and continuously interleave old and new as the network is further trained [Furlanello et al., 2016].”
Knowledge Distillation for Incremental Learning in Semantic Segmentation (2019, positive): “This idea was adapted in different ways in recent studies [1]– [5], [24], [25] to maintain the responses of the network on the old tasks whilst updating it with new training samples.”
Incremental Learning Techniques for Semantic Segmentation (2019, positive): “This idea was originally proposed in [4, 14] and then adapted in different ways in recent studies [5, 12, 18, 26, 30, 35, 37] to maintain stable the responses of the network on the old tasks whilst updating it with new training samples.”
Life-Long Disentangled Representation Learning with Cross-Domain Latent Homologies (2018, methodology): “The data-based approaches augment the training data on a new task with the data collected from the previous tasks, allowing for simultaneous multi-task learning on IID data [11, 45, 42, 33, 15].”
Lifelong Generative Modeling (2017, methodology): “(2015); Furlanello et al. (2016). The teacher contains a summary of all past distributions and can be perceived as a form of compressed memory.”
Memory-based Parameter Adaptation (2018, methodology): “…et al., 2017), where a penalty term is added to the loss for deviations far from previous weights, and learning without forgetting (Li & Hoiem, 2016; Furlanello et al., 2016), where distillation (Hinton et al., 2015) from previously trained models is used to keep old knowledge available.”
Continual Learning for Robotics (2019, methodology): “For example, the coreset can be used for distillation in [132] and in A-LTM (Active Long Term Memory Networks) [48] or to regularize the gradient when learning new tasks as in GEM (Gradient Episodic Memory) [92] and A-GEM (Averaged Gradient Episodic Memory) [23].”
Incremental Learning for Semantic Segmentation of Large-Scale Remote Sensing Data (2018, methodology): “The proposed methods in [41], [42] facilitate a similar distillation loss described in [40] to maintain performance on the previous tasks.”
A Strategy for an Uncompromising Incremental Learner (2017, background): “The authors Furlanello et al. develop a closely related procedure to in [3].”
Closed-Loop Memory GAN for Continual Learning (2018, methodology): “At each new task, these methods learn by presenting a network with both new images as well as replay of estimated or buffered old images, reverting the continual framework into a multi-task setting and thus alleviating forgetting [Furlanello et al., 2016].”
Continual Classification Learning Using Generative Models (2018, methodology): “The model has a student-teacher architecture (similar to that in distillation methods, [4], [2]), where the teacher contains a summary of all past distributions and is able to generate data from the previous tasks once we no longer have access to the original data.”
Knowledge Transfer with Jacobian Matching (2018, methodology): “(2016) and Furlanello et al. (2016). In this method, the student network is trained with a composite loss function involving two terms, one in each output branch.”