Continual Learning for Reliable In-Sequence Molecular Communication

Siddhant Setia; Junichi Suzuki; Tadashi Nakano

cs.LGOct 14, 2026

Continual Learning for Reliable In-Sequence Molecular Communication

Authors: Siddhant Setia, Junichi Suzuki, Tadashi Nakano

Organizations: Department of Computer Science, University of Massachusetts, Boston, Boston, MA, 02125, USA · Graduate School of Informatics, Osaka Metropolitan University, Osaka, Japan

Abstract

This paper considers reliable in-sequence molecular communication, where bio-nanomachines exchange molecule-encoded information in a specified order, and investigates a new simulation-based method to incrementally evaluate communication protocols by gradually increasing the complexity of experimental settings. The proposed method, called Affine Memory Replay (AMR), builds a continual learning (CL) model to estimate communication performance between bio-nanomachines. AMR stores past simulation samples and synthesizes additional replay samples through interpolation and extrapolation between past samples. Through incremental performance estimation tasks, it replays both past and synthetic samples to regularize model updates, helping the model adapt to new experimental settings while retaining knowledge from earlier ones. Experimental results show that AMR significantly outperforms existing CL methods in maintaining overall estimation accuracy across sequentially learned estimation tasks. It effectively balances memory stability and learning plasticity under moderate training cost.

Figures & tables

Fig. 1: Diffusive, Directional and Hybrid Transports with SW-ARQ in Molecular Communication

Fig. 2: Interaction between the Tx and the Rx

Method	Parameter	Value	Method	Parameter	Value
AMR	$\alpha$	1	CLeaR	$\alpha$	0.5
	$\beta$	0.5		$\lambda$	2
	Primary buffer size	50		Buffer size	50
	Secondary buffer size	50
	Synthesis range param $q$	0.1
ACLT	$\alpha$	0.2	A2ER	$\alpha$	1

TABLE I: Hyperparameter configuration

Parameter	Value
Size length of the environment	150 $\mu$ m
Diameter of Tx and Rx	5 $\mu$ m
Tx to Rx distance	10 to 90 $\mu$ m
Diameter of an information/ACK molecule	1 $\mu$ m
Diameter of a noise molecule	1 $\mu$ m
Number of noise molecules	0 to 10 5

TABLE II: Simulation Settings for Molecular Communication

Fig. 3: RTT Estimation Tasks (T1 to T12)

Method	$\mathcal{L}_{\text{plasticity}}$	$\mathcal{L}_{\text{stability}}$	$\mathcal{F}$
AMR	0.1069 $\pm$ 0.0112	0.1399 $\pm$ 0.0101	0.0330 $\pm$ 0.0138
ACLT	0.1073 $\pm$ 0.0116	0.1457 $\pm$ 0.0157 *!	0.0384 $\pm$ 0.0182
A2ER	0.1057 $\pm$ 0.0127	0.1463 $\pm$ 0.0158 *!	0.0406 $\pm$ 0.0173 *!
CLeaR	0.1075 $\pm$ 0.0115	0.1454 $\pm$ 0.0131 *!	0.0379 $\pm$ 0.0152

TABLE III: RTT Estimation Performance

Fig. 4: Change of forgetting ratio as the number of sequential tasks varies

Method	Total Training Time (sec)	Per-task Time (sec)
AMR	8.76 $\pm$ 1.50	0.73
ACLT	11.64 $\pm$ 1.02	0.97
A2ER	8.20 $\pm$ 1.07	0.68
CLeaR	7.39 $\pm$ 0.86	0.62

TABLE IV: Computational cost for training

Fig. 5: Tradeoff between Estimation Performance and Training Cost

$q$	$\mathcal{L}_{\text{plasticity}}$	$\mathcal{L}_{\text{stability}}$	$\mathcal{F}$
0.0	0.1053 $\pm$ 0.0090	0.1408 $\pm$ 0.0120	0.0355 $\pm$ 0.0155
0.05	0.1079 $\pm$ 0.0105	0.1415 $\pm$ 0.0080	0.0335 $\pm$ 0.0123
0.1	0.1069 $\pm$ 0.0113	0.1399 $\pm$ 0.0101	0.0330 $\pm$ 0.0138
0.2	0.1097 $\pm$ 0.0106	0.1405 $\pm$ 0.0105	0.0308 $\pm$ 0.0128
0.3	0.1076 $\pm$ 0.0115	0.1412 $\pm$ 0.0120	0.0336 $\pm$ 0.0162
1.5	0.1060 $\pm$ 0.0124	0.1416 $\pm$ 0.0126	0.0356 $\pm$ 0.0150

TABLE V: RTT Estimation Performance (AMR)

Fig. 6: Change in forgetting ratio as the number of tasks varies (AMR)

$q$	Total Training Time (sec)	Per-task Time (sec)
0.0	9.46 $\pm$ 1.73	0.79
0.05	10.29 $\pm$ 1.99	0.86
0.1	8.76 $\pm$ 1.50	0.73
0.2	8.80 $\pm$ 1.51	0.73
0.3	9.10 $\pm$ 1.71	0.76
1.5	10.04 $\pm$ 1.19	0.84

TABLE VI: Computational Cost for Training (AMR)

Fig. 7: Tradeoff between Estimation Performance and Training Cost (AMR)

Replay-Based Continual Learning

Replay-based continual learning preserves prior-task knowledge by revisiting stored or generated experience while adapting models to new data.

See all 132 papers

Continual Learning of Feedback-based Molecular Communication
May 1, 2026Siddhant Setia, Junichi Suzuki, Tadashi NakanoContinual Learning Replay-Based Continual Learning
Department of Computer Science, University of Massachusetts, Boston Boston, MA, 02125, USA · Graduate School of Informatics, Osaka Metropolitan University Osaka, Japan
Unsupervised Continual Learning for Amortized Bayesian Inference
Feb 26, 2026Aayush Mishra, Šimon Kucharský, Paul-Christian BürknerAmortized Inference Online Continual Learning
Department of Statistics, TU Dortmund University, Germany
Continuous Memory Machines
Oct 6, 2026Ciaran Regan, Kai Arulkumaran, Luke Darlow +3Memory-Augmented Neural Networks Recurrent Neural Networks
University of Tsukuba · Sakana AI

May 1, 2026J/K move · Enter open · S save

Continual Learning of Feedback-based Molecular Communication

Siddhant Setia, Junichi Suzuki, Tadashi Nakano

Department of Computer Science, University of Massachusetts, Boston Boston, MA, 02125, USA · Graduate School of Informatics, Osaka Metropolitan University Osaka, Japan

Continual Learning Replay-Based Continual Learning

Open paper AskPDF arXiv

This paper proposes and evaluates a new performance estimation method that leverages continual learning (CL) algorithms to carry out sequential simulation experiments for a feedback-based molecular communication protocol. As the protocol is sequentially examined in various experimental settings, the proposed CL-based performance estimators incrementally learn a series of unexperienced estimation tasks without compromising those that have been learned in the past. They are designed to work on a standard neural network architecture by customizing regularization and replay strategies in the loss function. Experimental results demonstrate that the proposed estimators can effectively learn on a continuous stream of simulation results and enhance the baseline neural network by improving estimation accuracy at a variety of computational costs. This paper's contribution is to establish the implications of CL in the field of molecular communication.

Abstract

Replay-Based Continual Learning

Explore similar work

Continual Learning of Feedback-based Molecular Communication

Unsupervised Continual Learning for Amortized Bayesian Inference

Continuous Memory Machines