ReConText3D is the first replay-memory framework for continual text-to-3D generation that prevents catastrophic forgetting on new textual categories while preserving quality on previously seen classes.
hub Canonical reference
Progressive Neural Networks
Canonical reference. 77% of citing Pith papers cite this work as background.
81
Pith papers citing it
Background
77% of classified citations
abstract
Learning to solve complex sequences of tasks--while both leveraging transfer and avoiding catastrophic forgetting--remains a key obstacle to achieving human-level intelligence. The progressive networks approach represents a step forward in this direction: they are immune to forgetting and can leverage prior knowledge via lateral connections to previously learned features. We evaluate this architecture extensively on a wide variety of reinforcement learning tasks (Atari and 3D maze games), and show that it outperforms common baselines based on pretraining and finetuning. Using a novel sensitivity measure, we demonstrate that transfer occurs at both low-level sensory and high-level control layers of the learned policy.
hub tools
citation-role summary
background 12
baseline 1
citation-polarity summary
claims ledger
- abstract Learning to solve complex sequences of tasks--while both leveraging transfer and avoiding catastrophic forgetting--remains a key obstacle to achieving human-level intelligence. The progressive networks approach represents a step forward in this direction: they are immune to forgetting and can leverage prior knowledge via lateral connections to previously learned features. We evaluate this architecture extensively on a wide variety of reinforcement learning tasks (Atari and 3D maze games), and show that it outperforms common baselines based on pretraining and finetuning. Using a novel sensitivi
co-cited works
representative citing papers
LIBERO is a new benchmark for lifelong robot learning that evaluates transfer of declarative, procedural, and mixed knowledge across 130 manipulation tasks with provided demonstration data.
CARLOS employs an aggregate deep neural network trained on progressively finer time grids with adaptive sampling to learn continuous-time exercise boundaries for optimal stopping, delivering higher values than discrete Bermudan methods.
MedCRP-CL discovers semantic modalities online via CRP from text prompts and maintains modality-specific LoRA adapters with intra-modality EWC, achieving 73.3% Dice and 4.1% forgetting on 16 tasks while using 6x fewer parameters than the best baseline.
Introduces replay-based continual learning with sequential invariance alignment to learn domain-invariant representations, outperforming baselines on generalization to unseen domains across six datasets in vision, medicine, manufacturing, and ecology.
MSRL represents trajectory segments as PSD matrices to prove additive composition properties and bootstrap value functions for better transfer, reaching 0.73 AUC versus 0.57-0.65 baselines.
KAN-CL cuts catastrophic forgetting by 88-93% on Split-CIFAR-10/5T and Split-CIFAR-100/10T by anchoring KAN parameters at per-knot granularity while matching baseline accuracy.
MIST fixes unreliable splits in streaming decision trees for class-incremental learning by replacing Hoeffding-style bounds with a K-independent McDiarmid radius on Gini, plus Bayesian parent-to-child inheritance and per-leaf quantile sketches.
A two-stage framework augments HOI data with dynamic priors and blends pre-trained dynamic motion and static interaction agents via a composer network to enable long-term dynamic human-object interactions with higher success rates and reduced training time.
Benchmark experiments in continual medical image segmentation reveal that no single method satisfies all clinical requirements, with replay-based approaches offering the best stability-plasticity trade-off while forward generalizability needs more attention.
A structure-aware VAE generates realistic FC matrices for replay, combined with multi-level knowledge distillation and hierarchical contextual bandit sampling, to enable continual fMRI-based brain disorder diagnosis across sequentially arriving multi-site data without catastrophic forgetting.
A hybrid SNN-LLM system uses learned spiking dynamics and lateral STDP propagation to trigger LLM actions without external prompts, producing the first autonomous action after 7 exchanges from a clean start.
SafeAdapt certifies a Rashomon set of safe policies from demonstration data and projects updates from arbitrary RL algorithms onto it to guarantee preservation of safety on source tasks.
SLE-FNO achieves zero forgetting and strong plasticity-stability balance in continual learning for FNO surrogate models of pulsatile blood flow by adding minimal single-layer extensions across four out-of-distribution tasks.
PRISM transfers RL policies zero-shot by aligning causally validated discrete concepts from agent encoders, achieving 69-76% win rates in Go 7x7 but random performance in Atari Breakout.
The paper offers a comprehensive survey and proposes a new taxonomy for continual learning strategies in VLMs and MLLMs to combat catastrophic forgetting beyond traditional methods.
LoRA adapters should be scaled by 1/sqrt(rank) rather than 1/rank to stabilize learning and enable effective use of higher ranks during fine-tuning of large language models.
Gato is a multi-modal, multi-task, multi-embodiment generalist policy using one transformer network to handle text, vision, games, and robotics tasks.
OpenAI Five achieved superhuman performance in Dota 2 by defeating the world champions using scaled self-play reinforcement learning.
NetTailor adapts CNN architecture for new tasks by assembling pre-trained universal blocks with task-specific layers, trained via activation mimicry and complexity penalties to match accuracy while reducing size for simpler tasks.
PROXYMIX learns a dynamic replay controller on a small proxy model and transfers it to a large target model, improving accuracy by 3.4 points and reducing forgetting by 3.5 points on LLaMA-3-8B continual tuning sequences.
Short-term data selectors in multi-stage LLM fine-tuning can slow future learning and increase forgetting, formalized as myopic selection with a proposed LHAS objective to address it.
Janus-LoRA uses gradient rectification via online subspace estimation and a decoupled margin loss to enforce parameter orthogonality and feature separation in LoRA-based continual learning, reporting new SOTA results.
PEAM is a parametric memory framework for Minecraft agents that internalizes experiences into a multimodal MoE-LoRA module using contrastive objectives on failures and a scale-free self-triggered consolidation mechanism.
citing papers explorer
-
Continuous-time Optimal Stopping through Deep Reinforcement Learning
CARLOS employs an aggregate deep neural network trained on progressively finer time grids with adaptive sampling to learn continuous-time exercise boundaries for optimal stopping, delivering higher values than discrete Bermudan methods.
-
Continual Learning of Domain-Invariant Representations
Introduces replay-based continual learning with sequential invariance alignment to learn domain-invariant representations, outperforming baselines on generalization to unseen domains across six datasets in vision, medicine, manufacturing, and ecology.
-
Matrix-Space Reinforcement Learning for Reusing Local Transition Geometry
MSRL represents trajectory segments as PSD matrices to prove additive composition properties and bootstrap value functions for better transfer, reaching 0.73 AUC versus 0.57-0.65 baselines.
-
KAN-CL: Per-Knot Importance Regularization for Continual Learning with Kolmogorov-Arnold Networks
KAN-CL cuts catastrophic forgetting by 88-93% on Split-CIFAR-10/5T and Split-CIFAR-100/10T by anchoring KAN parameters at per-knot granularity while matching baseline accuracy.
-
MIST: Reliable Streaming Decision Trees for Online Class-Incremental Learning via McDiarmid Bound
MIST fixes unreliable splits in streaming decision trees for class-incremental learning by replacing Hoeffding-style bounds with a K-independent McDiarmid radius on Gini, plus Bayesian parent-to-child inheritance and per-leaf quantile sketches.
-
SafeAdapt: Provably Safe Policy Updates in Deep Reinforcement Learning
SafeAdapt certifies a Rashomon set of safe policies from demonstration data and projects updates from arbitrary RL algorithms onto it to guarantee preservation of safety on source tasks.
-
SLE-FNO: Single-Layer Extensions for Task-Agnostic Continual Learning in Fourier Neural Operators
SLE-FNO achieves zero forgetting and strong plasticity-stability balance in continual learning for FNO surrogate models of pulsatile blood flow by adding minimal single-layer extensions across four out-of-distribution tasks.
-
Prism: Policy Reuse via Interpretable Strategy Mapping in Reinforcement Learning
PRISM transfers RL policies zero-shot by aligning causally validated discrete concepts from agent encoders, achieving 69-76% win rates in Go 7x7 but random performance in Atari Breakout.
-
Dota 2 with Large Scale Deep Reinforcement Learning
OpenAI Five achieved superhuman performance in Dota 2 by defeating the world champions using scaled self-play reinforcement learning.
-
Dynamic Proxy-Mixing: Transferring Replay Controllers from Small to Large Models for Continual Instruction Tuning
PROXYMIX learns a dynamic replay controller on a small proxy model and transfers it to a large target model, improving accuracy by 3.4 points and reducing forgetting by 3.5 points on LLaMA-3-8B continual tuning sequences.
-
The Long-Term Effects of Data Selection in LLM Fine-Tuning
Short-term data selectors in multi-stage LLM fine-tuning can slow future learning and increase forgetting, formalized as myopic selection with a proposed LHAS objective to address it.
-
Understanding Goal Generalisation in Sequential Reinforcement Learning
Empirical analysis of over 100 sequential RL training pipelines across 250+ OOD environments finds salient features drive generalization and early goals persist, with latent policy gradients simulating latent variable evolution to predict OOD behavior from training history.
-
TFGN: Task-Free, Replay-Free Continual Pre-Training Without Catastrophic Forgetting at LLM Scale
TFGN is an architectural overlay for transformers enabling task-free, replay-free continual pre-training across heterogeneous domains at LLM scale with near-zero backward transfer and high gradient orthogonality.
-
MoRe: Modular Representations for Principled Continual Representation Learning on Sequential Data
MoRe identifies modular structure in representations themselves to enable principled reuse, alignment, and expansion of modules during continual adaptation on sequential data.
-
Shortcut Solutions Learned by Transformers Impair Continual Compositional Reasoning
BERT learns shortcut solutions that impair generalization and forward transfer in continual LEGO, while ALBERT learns loop-like solutions for better performance, yet both fail at cross-experience composition, with ALBERT rescued by mixed-data training.
-
Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting
Sharpness-aware pretraining and related flat-minima interventions reduce catastrophic forgetting by up to 80% after post-training across 20M-150M models and by 31-40% at 1B scale.
-
NORACL: Neurogenesis for Oracle-free Resource-Adaptive Continual Learning
NORACL dynamically grows network capacity via neurogenesis-inspired signals to achieve oracle-level continual learning performance without pre-specifying architecture size.
-
Cortex-Inspired Continual Learning: Unsupervised Instantiation and Recovery of Functional Task Networks
FTN achieves near-zero forgetting on continual learning benchmarks by isolating task subnetworks via self-organizing binary masks generated through gradient descent, smoothing, and k-winner-take-all.
-
Information as Structural Alignment: A Dynamical Theory of Continual Learning
IBF achieves near-zero forgetting and positive backward transfer in continual learning by driving configurations toward coherence through motion and modification dynamics without storing raw data.
-
Adaptive Memory Crystallization for Autonomous AI Agent Learning in Dynamic Environments
AMC models memory consolidation via a Liquid-Glass-Crystal process governed by an SDE with proven convergence to a Beta distribution, yielding 34-43% better forward transfer and 67-80% less forgetting on standard continual RL benchmarks.
-
Causally Sufficient and Necessary Feature Expansion for Class-Incremental Learning
CPNS regularization with dual counterfactual generators mitigates intra-task and inter-task spurious correlations in class-incremental learning feature expansion.
-
CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
CrispEdit edits LLMs via low-curvature projections using Bregman divergence and K-FAC approximations, achieving high edit success with under 1% average capability degradation.
-
Robust Policy Optimization to Prevent Catastrophic Forgetting
FRPO applies a max-min robust optimization over KL-bounded policy neighborhoods during RLHF to reduce catastrophic forgetting of safety and accuracy under subsequent SFT or RL fine-tuning.
-
Routing-Based Continual Learning for Multimodal Large Language Models
Routing architecture for MLLMs enables continual learning with constant compute, matching multi-task learning performance and supporting cross-modal transfer.
-
A Survey of Continual Reinforcement Learning
The paper surveys CRL literature, proposes a taxonomy of methods into four categories based on knowledge storage and transfer, reviews metrics and benchmarks, and outlines challenges and future research directions.
-
No Forgetting Learning: Buffer-free Continual Learning Classification
NFL is a buffer-free continual learning framework that decomposes networks, applies stepwise freezing with knowledge distillation, and adds an auto-encoder in NFL+ to match replay-based performance on image benchmarks while using only 2.53% of the memory.
-
Attentive Multi-Task Deep Reinforcement Learning
Attention mechanism dynamically groups task knowledge at state granularity in multi-task DRL to enable positive transfer and avoid negative transfer, matching or exceeding prior methods with fewer parameters.
-
Lifelong Learning Starting From Zero
A blank-slate neural network grows via expansion, generalization, forgetting, and backpropagation for lifelong learning with claimed gains in accuracy, efficiency, and versatility.
-
Continual Reinforcement Learning with Diversity Exploration and Adversarial Self-Correction
CDAN framework uses diversity exploration and adversarial self-correction for continual RL in continuous control, evaluated on new CAM environment with NSD metric showing 18.35% NSD improvement over baseline.
-
CRMA: A Spectrally-Bounded Backbone for Modular Continual Fine-Tuning of LLMs
CRMA adds a spectrally bounded residual adapter backbone to modular continual fine-tuning of LLMs, achieving near-zero loss drift and positive backward transfer on Mistral-7B across domains.
-
TRACER: Persistent Regularization for Robust Multimodal Finetuning
TRACER applies weighted moving average distillation in contrastive finetuning of multimodal models to retain pretrained knowledge and boost out-of-distribution accuracy.
-
Tunable MAGMAX: Preference-Aware Model Merging for Continual Learning
Tunable MAGMAX adds a tunable preference vector to model merging for continual learning, enabling automatic adaptation to target environments using small amounts of data while maintaining or improving task-wise performance.
-
CP-MoE: Consistency-Preserving Mixture-of-Experts for Continual Learning
CP-MoE uses a transient expert, consistency-preserving routing bias, and guided regularization to reduce catastrophic forgetting in MoE-based LLMs and VLMs while preserving cross-task transfer, reporting SOTA on SuperNI and gains on VQA v2.
-
FLAME: Adaptive Mixture-of-Experts for Continual Multimodal Multi-Task Learning
FLAME is an MoE architecture using modality-specific routers and low-rank compression of expert knowledge to support efficient continual multimodal multi-task learning while reducing catastrophic forgetting.
-
Learning Material-Aware Hamiltonian Risk Fields for Safe Navigation
A learned context-energy term in port-Hamiltonian policies creates selective risk navigation that activates evasive forces only when safer paths are available.
-
A Domain Incremental Continual Learning Benchmark for ICU Time Series Model Transportability
Proposes a domain incremental continual learning benchmark for ICU time series model transportability across US regions and evaluates data replay and EWC methods.
-
Task Switching Without Forgetting via Proximal Decoupling
Operator splitting separates task optimization from proximal stability enforcement to achieve forgetting-free continual learning with SOTA benchmark results.
-
Neural Computers
Neural Computers are introduced as a new machine form where computation, memory, and I/O are unified in a learned runtime state, with initial video-model experiments showing acquisition of basic interface primitives from traces.
-
Growing Transformers: Modular Composition and Layer-wise Expansion on a Frozen Substrate
Demonstrates that Transformers can continue learning when grown modularly above a frozen minimal token interface under a fixed active-parameter budget, with reported viability in 9-layer and 16-layer experiments.
-
Beneficial perturbation network for continual learning
BPN adds task-specific beneficial perturbations as biases to neural networks to overcome catastrophic forgetting without storing prior data or expanding the network substantially.
-
Revitalizing the Beginning: Avoiding Storage Dependency for Model Merging in Continual Learning
The paper proposes Trajectory Regularized Merging (TRM) to enable storage-free model merging in continual learning by optimizing in an augmented trajectory subspace with task alignment, prediction consistency, and gradient responsiveness objectives, claiming SOTA results.
-
MPCS: Neuroplastic Continual Learning via Multi-Component Plasticity and Topology-Aware EWC
MPCS integrates eleven plasticity mechanisms and reaches a Normalized Efficiency Score of 94.2 on a 31-task benchmark, with ablations showing that removing EWC and Hebbian updates yields higher performance at lower cost.
-
Self-Distillation as a Performance Recovery Mechanism for LLMs: Counteracting Compression and Catastrophic Forgetting
Self-distillation fine-tuning recovers LLM capabilities by aligning the student's high-dimensional hidden-layer manifold with the teacher's, as quantified by CKA correlation with performance gains.
-
Incremental Concept Learning via Online Generative Memory Recall
Pseudo-rehearsal method with cGAN-generated old-concept samples, balanced online recall, and concept contrastive loss for class-incremental learning on MNIST, Fashion-MNIST and SVHN.
- On the Stability of Growth in Structural Plasticity
- Little by Little: Continual Learning via Incremental Mixture of Rank-1 Associative Memory Experts