Soft Merging of Experts with Adaptive Routing

• URL: http://arxiv.org/abs/2306.03745v2
• Date: Mon, 13 May 2024 16:20:29 GMT
• Title: Soft Merging of Experts with Adaptive Routing
• Abstract summary: We introduce Soft Merging of Experts with Adaptive Routing (SMEAR) SMEAR avoids discrete routing by using a single "merged" expert constructed via a weighted average of all of the experts' parameters. We empirically validate that models using SMEAR outperform models that route based on metadata or learn sparse routing through gradient estimation.
• Score: 38.962451264172856
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparsely activated neural networks with conditional computation learn to route their inputs through different "expert" subnetworks, providing a form of modularity that densely activated models lack. Despite their possible benefits, models with learned routing often underperform their parameter-matched densely activated counterparts as well as models that use non-learned heuristic routing strategies. In this paper, we hypothesize that these shortcomings stem from the gradient estimation techniques used to train sparsely activated models that use non-differentiable discrete routing decisions. To address this issue, we introduce Soft Merging of Experts with Adaptive Routing (SMEAR), which avoids discrete routing by using a single "merged" expert constructed via a weighted average of all of the experts' parameters. By routing activations through a single merged expert, SMEAR does not incur a significant increase in computational costs and enables standard gradient-based training. We empirically validate that models using SMEAR outperform models that route based on metadata or learn sparse routing through gradient estimation. Furthermore, we provide qualitative analysis demonstrating that the experts learned via SMEAR exhibit a significant amount of specialization. All of the code used in our experiments is publicly available.

Related papers

• When Model Merging Breaks Routing: Training-Free Calibration for MoE [67.27839627141972]
  Hessian-Aware Router (HARC) is a training-free framework that leverages second-order curvature information to realign the merged router.<n> Experiments on mathematical reasoning and code generation tasks show that HARC effectively mitigates routing breakdown across diverse MoE merging baselines.
  arXiv  Detail & Related papers  (2026-06-02T09:33:33Z)
• Learning Multi-Modal Trajectory Policies for Data-Efficient Robotic Manipulation [21.99077052779724]
  MATE (Multi-Modal MoE) is a novel trajectory prediction framework built upon MoE.<n>We introduce a Multi-Modal MoE architecture to achieve fine-grained sub-modal cosine, and design a cross-modal router for stable and scale-invariant expert assignment.<n>Experiments on robotic ping-pong show that our MATE consistently outperforms prior work under data scarcity.
  arXiv  Detail & Related papers  (2026-05-31T06:35:10Z)
• Multi-Domain Learning with Global Expert Mapping [102.62297074508147]
  Mixture-of-Experts (MoE) models provide a scalable solution by routing inputs to specializedworks (experts)<n>We propose GEM (Global Expert Mapping), a planner-compiler framework that replaces the learned router with a global scheduler.<n>Our planner, based on linear programming relaxation, computes a fractional assignment of datasets to experts, while the compiler applies hierarchical rounding to convert this soft plan into a deterministic, capacity-aware mapping.<n>Experiments show that GEM-DINO achieves state-of-the-art performance on the UODB benchmark, with notable gains on underrepresented datasets and solves task interference in few
  arXiv  Detail & Related papers  (2026-04-20T21:09:34Z)
• MoE Routing Testbed: Studying Expert Specialization and Routing Behavior at Small Scale [12.078286327859425]
  We propose the MoE Routing Testbed, a setup that gives clearer visibility into routing dynamics at small scale.<n>The testbed pairs a data mix with clearly distinguishable domains with a reference router that prescribes ideal routing.<n>We show that balancing scope is the crucial factor that allows specialization while maintaining high expert utilization.
  arXiv  Detail & Related papers  (2026-04-08T12:47:57Z)
• CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training [48.70704477452434]
  We introduce textbfCapTrack, a capability-centric framework for analyzing forgetting in large language models.<n>We conduct a large-scale empirical study across post-training algorithms, domains, and model families.<n>We find that forgetting extends beyond parametric knowledge, with pronounced drift in robustness and default behaviors.
  arXiv  Detail & Related papers  (2026-02-19T09:46:24Z)
• Expert Divergence Learning for MoE-based Language Models [21.857507958048696]
  We introduce Expert Divergence Learning, a novel pre-training strategy that explicitly encourages functional specialization among experts.<n>Our method incorporates a label-driven auxiliary loss that leverages domain labels inherent in pre-training corpora.<n>We validate our approach by pre-training MoE models of up to 15 billion parameters from scratch.
  arXiv  Detail & Related papers  (2026-02-10T06:58:38Z)
• Guided by the Experts: Provable Feature Learning Dynamic of Soft-Routed Mixture-of-Experts [11.437368205968573]
  This paper advances MoE theory by providing convergence guarantees for joint training of soft-routed MoE models with non-linear routers and experts.<n>We show that a post-training pruning can effectively eliminate redundant neurons, followed by a provably convergent fine-tuning process that reaches global optimality.
  arXiv  Detail & Related papers  (2025-10-08T16:40:31Z)
• LLM Routing with Dueling Feedback [49.67815163970033]
  We study the problem of selecting the best model for each query while balancing user satisfaction, model expertise, and inference cost.<n>We formulate routing as contextual dueling bandits, learning from pairwise preference feedback rather than absolute scores.<n>We introduce Category-Calibrated Fine-Tuning (CCFT), a representation-learning method that derives model embeddings from offline data using contrastive fine-tuning with categorical weighting.
  arXiv  Detail & Related papers  (2025-10-01T12:52:25Z)
• LD-MoLE: Learnable Dynamic Routing for Mixture of LoRA Experts [24.0422448103907]
  We propose LD-MoLE, a Learnable Dynamic routing mechanism for Mixture of LoRA Experts.<n>Our design allows the model to adaptively determine the number of experts to activate for each token at different layers.<n>Our method achieves superior performance, but also demonstrates the ability to learn token-dependent and layer-wise expert allocation.
  arXiv  Detail & Related papers  (2025-09-30T02:38:10Z)
• MoE-CE: Enhancing Generalization for Deep Learning based Channel Estimation via a Mixture-of-Experts Framework [34.580240531578106]
  MoE-CE is a flexible mixture-of-experts framework designed to enhance the generalization capability of DL-based CE methods.<n>We show that MoE-CE consistently outperforms conventional DL approaches, achieving significant performance gains while maintaining efficiency.
  arXiv  Detail & Related papers  (2025-09-19T13:23:08Z)
• Advancing Expert Specialization for Better MoE [22.570561334474252]
  Mixture-of-Experts (MoE) models enable efficient scaling of large language models (LLMs) by activating only a subset of experts per input.<n>We observe that the commonly used auxiliary load balancing loss often leads to expert overlap and overly uniform routing.<n>We propose a simple yet effective solution that introduces two complementary objectives.
  arXiv  Detail & Related papers  (2025-05-28T13:09:47Z)
• Improving Routing in Sparse Mixture of Experts with Graph of Tokens [32.46693871593765]
  We unveil the limitation of Sparse Mixture of Experts (SMoE) through the perspective of the probabilistic graphical model (PGM)<n>We propose the novel Similarity-Aware (S)MoE, which considers interactions between tokens during expert selection.<n>We empirically validate our models on various tasks and domains, showing significant improvements in reducing routing fluctuations.
  arXiv  Detail & Related papers  (2025-05-01T18:44:20Z)
• Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts [33.39800923804871]
  We introduce Race-DiT, a novel MoE model for diffusion transformers with a flexible routing strategy, Expert Race. By allowing tokens and experts to compete together and select the top candidates, the model learns to dynamically assign experts to critical tokens.
  arXiv  Detail & Related papers  (2025-03-20T11:45:08Z)
• DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs [70.91804882618243]
  This paper proposes DSMoE, a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks. We implement adaptive expert routing using sigmoid activation and straight-through estimators, enabling tokens to flexibly access different aspects of model knowledge. Experiments on LLaMA models demonstrate that under equivalent computational constraints, DSMoE achieves superior performance compared to existing pruning and MoE approaches.
  arXiv  Detail & Related papers  (2025-02-18T02:37:26Z)
• Revisiting SMoE Language Models by Evaluating Inefficiencies with Task Specific Expert Pruning [78.72226641279863]
  Sparse Mixture of Expert (SMoE) models have emerged as a scalable alternative to dense models in language modeling. Our research explores task-specific model pruning to inform decisions about designing SMoE architectures. We introduce an adaptive task-aware pruning technique UNCURL to reduce the number of experts per MoE layer in an offline manner post-training.
  arXiv  Detail & Related papers  (2024-09-02T22:35:03Z)
• Performance Characterization of Expert Router for Scalable LLM Inference [0.4726677580049183]
  Large Language Models (LLMs) have experienced widespread adoption across scientific and industrial domains. deploying and serving these models at scale with optimal throughput and latency remains a significant challenge. This paper introduces Expert Router, a scalable routing architecture that directs to specialized expert models.
  arXiv  Detail & Related papers  (2024-04-22T16:33:42Z)
• Domain Generalization Guided by Gradient Signal to Noise Ratio of Parameters [69.24377241408851]
  Overfitting to the source domain is a common issue in gradient-based training of deep neural networks. We propose to base the selection on gradient-signal-to-noise ratio (GSNR) of network's parameters.
  arXiv  Detail & Related papers  (2023-10-11T10:21:34Z)
• Phantom Embeddings: Using Embedding Space for Model Regularization in Deep Neural Networks [12.293294756969477]
  The strength of machine learning models stems from their ability to learn complex function approximations from data. The complex models tend to memorize the training data, which results in poor regularization performance on test data. We present a novel approach to regularize the models by leveraging the information-rich latent embeddings and their high intra-class correlation.
  arXiv  Detail & Related papers  (2023-04-14T17:15:54Z)
• Model-Based Deep Learning: On the Intersection of Deep Learning and Optimization [101.32332941117271]
  Decision making algorithms are used in a multitude of different applications. Deep learning approaches that use highly parametric architectures tuned from data without relying on mathematical models are becoming increasingly popular. Model-based optimization and data-centric deep learning are often considered to be distinct disciplines.
  arXiv  Detail & Related papers  (2022-05-05T13:40:08Z)
• Distributionally Robust Models with Parametric Likelihood Ratios [123.05074253513935]
  Three simple ideas allow us to train models with DRO using a broader class of parametric likelihood ratios. We find that models trained with the resulting parametric adversaries are consistently more robust to subpopulation shifts when compared to other DRO approaches.
  arXiv  Detail & Related papers  (2022-04-13T12:43:12Z)
• Interpretable AI-based Large-scale 3D Pathloss Prediction Model for enabling Emerging Self-Driving Networks [3.710841042000923]
  We propose a Machine Learning-based model that leverages novel key predictors for estimating pathloss. By quantitatively evaluating the ability of various ML algorithms in terms of predictive, generalization and computational performance, our results show that Light Gradient Boosting Machine (LightGBM) algorithm overall outperforms others.
  arXiv  Detail & Related papers  (2022-01-30T19:50:16Z)
• Taming Sparsely Activated Transformer with Stochastic Experts [76.0711573018493]
  Sparsely activated models (SAMs) can easily scale to have outrageously large amounts of parameters without significant increase in computational cost. In this paper, we propose a new expert-based model, THOR (Transformer witH StOchastic ExpeRts) Unlike classic expert-based models, such as the Switch Transformer, experts in THOR are randomly activated for each input during training and inference.
  arXiv  Detail & Related papers  (2021-10-08T17:15:47Z)
• Last Layer Marginal Likelihood for Invariance Learning [12.00078928875924]
  We introduce a new lower bound to the marginal likelihood, which allows us to perform inference for a larger class of likelihood functions. We work towards bringing this approach to neural networks by using an architecture with a Gaussian process in the last layer.
  arXiv  Detail & Related papers  (2021-06-14T15:40:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.