Lifelong Generative Modeling

By Jason Ramapuram, Magda Gregorová, Alexandros Kalousis. 2017

134 citations recorded by the source corpus.

Abstract Lifelong learning is the problem of learning multiple consecutive tasks in a sequential manner, where knowledge gained from previous tasks is retained and used to aid future learning over the lifetime of the learner. It is essential towards the development of intelligent machines that can adapt to their surroundings. In this work we focus on a lifelong learning approach to unsupervised generative modeling, where we continuously incorporate newly observed distributions into a learned model. We do so through a student-teacher Variational Autoencoder architecture which allows us to learn and preserve all the distributions seen so far, without the need to retain the past data nor the past models. Through the introduction of a novel cross-model regularizer, inspired by a Bayesian update rule, the student model leverages the information learned by the teacher, which acts as a probabilistic knowledge store. The regularizer reduces the effect of catastrophic interference that appears when we learn over sequences of distributions. We validate our model’s performance on sequential variants of MNIST, FashionMNIST, PermutedMNIST, SVHN and Celeb-A and demonstrate that our model mitigates the effects of catastrophic interference faced by neural networks in sequential learning scenarios.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Generative Models, Meta & Continual Learning

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 Computer Vision and Pattern Recognition

In Generative Models

In Meta & Continual Learning

Citations in the corpus

Accelerating Online Reinforcement Learning with Offline Datasets (2020, background): “Training density models online (in the “streaming” setting) is a challenging research problem [20], made more difficult by a potentially complex multi-modal behavior distribution induced by the mixture of online and offline data.”
NISPA: Neuro-Inspired Stability-Plasticity Adaptation for Continual Learning in Sparse Networks (2022, methodology): “These methods store (Rebufﬁ et al., 2017; Rolnick et al., 2019; Isele & Cosgun, 2018) or generate (Shin et al., 2017; Atkinson et al., 2018; Ramapuram et al., 2020) examples of previous tasks.”
The Challenges of Continuous Self-Supervised Learning (2022, background): “However, recent works [1, 63, 66, 82] have attempted to address a sub-problem of ours, i.”
IIRC: Incremental Implicitly-Refined Classification (2020, background): “Other works include LGM [37], IL2M [6], BIC [51], and ER [39].”
Learning latent representations across multiple data domains using Lifelong VAEGAN (2020, methodology): “ to load all previously learnt data for the generation task. Approaches employing both generative and inference mechanisms are based on the VAE framework Accepted as a conference paper at ECCV 2020 3 [1,50]. However, these approaches have degenerating performance when learning high-dimensional data,”
Online Continual Learning with Maximally Interfered Retrieval (2019, background): “Continual learning has also been studied recently for the case of learning generative models [29, 22].”
Life-Long Disentangled Representation Learning with Cross-Domain Latent Homologies (2018, methodology): “It is also worth noting that VASE outperforms the “SD-[41]” condition, which is similar to the only other VAE-based approach to continual learning that we are aware of, see [41].”
Scalable Recollections for Continual Lifelong Learning (2017, background): “Some approaches to lifelong learning assume that a task is a set of examples chosen from the same distribution (Rusu et al., 2016; Fernando et al., 2017; Shin et al., 2017a; Ramapuram et al., 2017; Al-Shedivat et al., 2017; Lee et al., 2018).”
Continual learning: A comparative study on how to defy forgetting in classification tasks (2019, methodology): “Rehearsal iCaRL [18] ER [44] SER [45] TEM [46] DGR [14] PR [47] CCLUGM [48] LGM [49] Constrained GEM [50] A-GEM [8] GSS [43] Regularization-basedmethods Prior-focused EWC [28] IMM [29] SI [51] R-EWC [52] MAS [15] [16] Data-focused LwF [53] LFL [54] EBLL [11] DMC [55] Parameter isolation methods”
Generative Models from the perspective of Continual Learning (2018, methodology): “A second method experimented on VAEs is to use a studentteacher method where the student learns the current task while the teacher retains knowledge [21].”
Continual Learning for Robotics (2019, neutral): “, 2018b), variational auto-encoders (Nguyen et al., 2017; Ramapuram et al., 2017; Achille et al., 2018; Farquhar and Gal, 2018; Lesort et al., 2018b) and standard auto-encoders (Triki et al.”
A Continual Learning Survey: Defying Forgetting in Classification Tasks (2019, methodology): “Rehearsal iCaRL [16] ER[49]SER[50] TEM[51]CoPE[33] DGR[12]PR[52] CCLUGM [53] LGM[54] Constrained GEM [55] A-GEM[6]GSS[48] Regularization-basedmethods Prior-focused EWC[27]IMM[28]SI[56] R-EWC[57]MAS[13] [14] Data-focused LwF[58]LFL[59]EBLL[9] DMC[60] Parameter isolation methods”
Lifelong Teacher-Student Network Learning (2021, methodology): “Ramapuram et al. [33] proposed the Lifelong Generative Modeling (LGM) which employs VAEs for two networks working in tandem: a Teacher and a Student.”
DDGR: Continual Learning with Deep Diffusion-based Generative Replay (2023, methodology): “Classical GR methods use VAE or GAN as the generator for CL (Wu et al., 2018; Achille et al., 2018; Zhai et al., 2019; Wu et al., 2019; Ramapuram et al., 2020; Mundt et al., 2022).”
Lifelong learning of interpretable image representations (2020, methodology): “FID is evaluated on both CelebA and 3D-Chair images, and the numerical results are provided in the bar-plot from Figure 8, where we compare the proposed model with other lifelong learning approaches, such as LGAN [27] and LGM [28].”
Online Learned Continual Compression with Stacked Quantization Module (2019, background): “Although CL has been studied in generative modeling (Ramapuram et al., 2017; Lesort et al., 2018; Zhai et al., 2019; Lesort et al., 2019) and reinforcement learning (Kirkpatrick et al., 2017; Fernando et al., 2017; Riemer et al., 2018), supervised learning is still the standard for evaluation of…”
Continual Variational Autoencoder Learning via Online Cooperative Memorization (2022, background): “The first would usually train a generator [2,44,49], or store a few past ar X iv :2 20 7.”
Lifelong Twin Generative Adversarial Networks (2021, methodology): “Dataset LT-GANs CURL [5] LGM [23] MNIST 878.”