Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Chisme: Fully Decentralized Differentiated Deep Learning for IoT Intelligence

Harikrishna Kuttivelil, Katia Obraczka

TL;DR

Chisme tackles the challenge of learning at the network edge under data heterogeneity and intermittent connectivity by introducing a fully decentralized, differentiation-aware learning protocol. It builds a cosine-similarity-based data affinity heuristic from incoming model deltas to modulate how neighboring updates are merged, replacing uniform collaboration with selective, affinity-guided aggregation. The approach extends gossip learning with a novel combined influence coefficient $\eta_k$ and maintains a per-neighbor experience map to support asynchronous operation and memory efficiency. Empirical results on image and time-series tasks under varied network conditions show that Chisme achieves faster convergence, lower final loss, and reduced inter-client disparity compared to state-of-the-art centralized and decentralized baselines. This work advances practical edge intelligence by enabling robust, personalized collaboration without centralized infrastructure.

Abstract

As end-user device capability increases and demand for intelligent services at the Internet's edge rise, distributed learning has emerged as a key enabling technology. Existing approaches like federated learning (FL) and decentralized FL (DFL) enable distributed learning among clients, while gossip learning (GL) approaches have emerged to address the potential challenges in resource-constrained, connectivity-challenged infrastructure-less environments. However, most distributed learning approaches assume largely homogeneous data distributions and may not consider or exploit the heterogeneity of clients and their underlying data distributions. This paper introduces Chisme, a novel fully decentralized distributed learning algorithm designed to address the challenges of implementing robust intelligence in network edge contexts characterized by heterogeneous data distributions, episodic connectivity, and sparse network infrastructure. Chisme leverages cosine similarity-based data affinity heuristics calculated from received model exchanges to inform how much influence received models have when merging into the local model. By doing so, it facilitates stronger merging influence between clients with more similar model learning progressions, enabling clients to strategically balance between broader collaboration to build more general knowledge and more selective collaboration to build specific knowledge. We evaluate Chisme against contemporary approaches using image recognition and time-series prediction scenarios while considering different network connectivity conditions, representative of real-world distributed intelligent systems. Our experiments demonstrate that Chisme outperforms state-of-the-art edge intelligence approaches in almost every case -- clients using Chisme exhibit faster training convergence, lower final loss after training, and lower performance disparity between clients.

Chisme: Fully Decentralized Differentiated Deep Learning for IoT Intelligence

TL;DR

Chisme tackles the challenge of learning at the network edge under data heterogeneity and intermittent connectivity by introducing a fully decentralized, differentiation-aware learning protocol. It builds a cosine-similarity-based data affinity heuristic from incoming model deltas to modulate how neighboring updates are merged, replacing uniform collaboration with selective, affinity-guided aggregation. The approach extends gossip learning with a novel combined influence coefficient and maintains a per-neighbor experience map to support asynchronous operation and memory efficiency. Empirical results on image and time-series tasks under varied network conditions show that Chisme achieves faster convergence, lower final loss, and reduced inter-client disparity compared to state-of-the-art centralized and decentralized baselines. This work advances practical edge intelligence by enabling robust, personalized collaboration without centralized infrastructure.

Abstract

As end-user device capability increases and demand for intelligent services at the Internet's edge rise, distributed learning has emerged as a key enabling technology. Existing approaches like federated learning (FL) and decentralized FL (DFL) enable distributed learning among clients, while gossip learning (GL) approaches have emerged to address the potential challenges in resource-constrained, connectivity-challenged infrastructure-less environments. However, most distributed learning approaches assume largely homogeneous data distributions and may not consider or exploit the heterogeneity of clients and their underlying data distributions. This paper introduces Chisme, a novel fully decentralized distributed learning algorithm designed to address the challenges of implementing robust intelligence in network edge contexts characterized by heterogeneous data distributions, episodic connectivity, and sparse network infrastructure. Chisme leverages cosine similarity-based data affinity heuristics calculated from received model exchanges to inform how much influence received models have when merging into the local model. By doing so, it facilitates stronger merging influence between clients with more similar model learning progressions, enabling clients to strategically balance between broader collaboration to build more general knowledge and more selective collaboration to build specific knowledge. We evaluate Chisme against contemporary approaches using image recognition and time-series prediction scenarios while considering different network connectivity conditions, representative of real-world distributed intelligent systems. Our experiments demonstrate that Chisme outperforms state-of-the-art edge intelligence approaches in almost every case -- clients using Chisme exhibit faster training convergence, lower final loss after training, and lower performance disparity between clients.
Paper Structure (11 sections, 8 equations, 4 figures, 2 tables)

This paper contains 11 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background & Related Work
  3. Chisme: Collaborative Heterogeneity-aware Intelligence via Similarity-based Model Exchanges
  4. Fully Decentralized Collaborative Learning
  5. Inferring Data Similarity through Cosine Similarity
  6. Cosine Similarity-based Decentralized Federated Learning
  7. Chisme: Fully Decentralized Differentiated Deep Learning
  8. Results
  9. Methodology
  10. Discussion
  11. Conclusion

Figures (4)

  • Figure 1: Simplified representation of non-differentiated collaborative learning.
  • Figure 2: Loss mean and standard deviation ($\sigma$) across clients in the FEMNIST handwritten digit recognition scenario.
  • Figure 3: Loss mean and standard deviation ($\sigma$) across clients in the artificially incongruent, label-swapped MNIST handwritten digit recognition scenario.
  • Figure 4: Loss mean and standard deviation ($\sigma$) across clients in the Irish Weather Station multi-feature weather prediction scenario.