Sieve dynamically schedules MoE experts across GPU and PIM hardware to handle bimodal token distributions, achieving 1.3x to 1.6x gains in throughput and interactivity over static prior PIM systems on three large models.
citation dossier
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
The capacity of a neural network to absorb information is limited by its number of parameters. Conditional computation, where parts of the network are active on a per-example basis, has been proposed in theory as a way of dramatically increasing model capacity without a proportional increase in computation. In practice, however, there are significant algorithmic and performance challenges. In this work, we address these challenges and finally realize the promise of conditional computation, achieving greater than 1000x improvements in model capacity with only minor losses in computational efficiency on modern GPU clusters. We introduce a Sparsely-Gated Mixture-of-Experts layer (MoE), consisting of up to thousands of feed-forward sub-networks. A trainable gating network determines a sparse combination of these experts to use for each example. We apply the MoE to the tasks of language modeling and machine translation, where model capacity is critical for absorbing the vast quantities of knowledge available in the training corpora. We present model architectures in which a MoE with up to 137 billion parameters is applied convolutionally between stacked LSTM layers. On large language modeling and machine translation benchmarks, these models achieve significantly better results than state-of-the-art at lower computational cost.
116Pith papers citing it
120reference links
cs.LGtop field · 38 papers
UNVERDICTEDtop verdict bucket · 105 papers
why this work matters in Pith
Pith has found this work in 116 reviewed papers. Its strongest current cluster is cs.LG (38 papers). The largest review-status bucket among citing papers is UNVERDICTED (105 papers). For highly cited works, this page shows a dossier first and a bounded explorer second; it never tries to render every citing paper at once.
representative citing papers
ArgBench unifies 33 existing datasets into a standardized benchmark for testing LLMs across 46 argumentation tasks and analyzes the impact of prompting techniques and model factors on performance.
GPT-3 shows that scaling an autoregressive language model to 175 billion parameters enables strong few-shot performance across diverse NLP tasks via in-context prompting without fine-tuning.
Routers in SMoE models form geometric alignments with their experts through shared gradient directions, enabling effective specialization that auxiliary load-balancing losses tend to disrupt.
EEP makes wide expert-parallel MoE serving survive single-rank failures with an 11s recovery pause, 8s reintegration pause, and throughput restored to 95% of pre-fault level within 52s while staying within 4.4% of a fixed-membership baseline in steady state.
Test-Time Speculation adapts draft models online via target-model verifications to sustain high acceptance lengths during long LLM generations.
Dooly reduces LLM inference profiling costs by 56.4% via configuration-agnostic taint-based labeling and selective database reuse, delivering simulation accuracy within 5% MAPE for TTFT and 8% for TPOT across 12 models.
SDG-MoE introduces learned signed interaction graphs and disagreement-gated deliberation among experts in MoE architectures, yielding 19.8% better validation perplexity than the strongest baseline.
DTSemNet gives an exact, invertible neural-network encoding of hard oblique decision trees that supports direct gradient training for both classification and regression without probabilistic softening or quantized estimators.
MISA routes to a small subset of indexer heads via block statistics, matching full DSA performance on LongBench with 4-8x fewer heads and 3.82x speedup while recovering over 92% of selected tokens.
SplatWeaver dynamically allocates Gaussian primitives via cardinality experts and pixel-level routing guided by high-frequency cues for improved generalizable novel view synthesis.
Standard top-k routers in MoE language models often select suboptimal routes for difficult tokens, and updating only the final router layer raises pass@K on AIME and HMMT benchmarks across multiple models.
TENNOR enables efficient private training of wide neural networks in TEEs by recasting sparsification as doubly oblivious LSH retrievals and introducing MP-WTA to cut hash table memory by 50x while preserving accuracy.
MoLF routes updates between full fine-tuning and LoRA at the optimizer level to match or exceed the better of either static method, with an efficient LoRA-only variant outperforming prior adaptive approaches.
CEA assembles per-token low-rank residual updates via dense affinities over hyper-adapter-generated components to improve all-in-one image restoration on spatially non-uniform degradations.
A buffer-free MoE dispatch and combine method on Ascend hardware with pooled HBM cuts intermediate relay overhead via direct expert window access.
Misrouter enables input-only attacks on MoE LLMs by optimizing queries on open-source surrogates to route toward weakly aligned experts and transferring them to public APIs.
Adding temporal memory via LIF, precision-weighted gating, and anticipatory prediction to MoE routers recovers effective expert selection at distribution transitions, with ablation confirming a super-additive beta-ant interaction.
RouteHijack is a routing-aware jailbreak that identifies safety-critical experts via activation contrast and optimizes suffixes to suppress them, reaching 69.3% average attack success rate on seven MoE LLMs with strong transfer to variants and VLMs.
MASCing uses an LSTM surrogate and optimized steering masks to enable flexible, inference-time control over MoE expert routing for safety objectives, improving jailbreak defense and content generation success rates substantially across multiple models.
DMEP prunes experts module-by-module in LoRA-MoE and removes load balancing after pruning, cutting trainable parameters 35-43% and raising throughput ~10% while matching or exceeding uniform baselines on reasoning tasks.
Hallo-Live achieves 20.38 FPS real-time text-to-audio-video avatar generation with 0.94s latency using asynchronous dual-stream diffusion and HP-DMD preference distillation, matching teacher model quality at 16x higher throughput.
Expert upcycling duplicates experts in an existing MoE checkpoint and continues pre-training to match fixed-size baseline performance with 32% less compute.
LLM planners for robots often produce dangerous plans even when planning succeeds, with safety awareness staying flat as model scale improves planning ability.
citing papers explorer
-
Sieve: Dynamic Expert-Aware PIM Acceleration for Evolving Mixture-of-Experts Models
Sieve dynamically schedules MoE experts across GPU and PIM hardware to handle bimodal token distributions, achieving 1.3x to 1.6x gains in throughput and interactivity over static prior PIM systems on three large models.
-
ArgBench: Benchmarking LLMs on Computational Argumentation Tasks
ArgBench unifies 33 existing datasets into a standardized benchmark for testing LLMs across 46 argumentation tasks and analyzes the impact of prompting techniques and model factors on performance.
-
Language Models are Few-Shot Learners
GPT-3 shows that scaling an autoregressive language model to 175 billion parameters enables strong few-shot performance across diverse NLP tasks via in-context prompting without fine-tuning.
-
Routers Learn the Geometry of Their Experts: Geometric Coupling in Sparse Mixture-of-Experts
Routers in SMoE models form geometric alignments with their experts through shared gradient directions, enabling effective specialization that auxiliary load-balancing losses tend to disrupt.
-
Surviving Partial Rank Failures in Wide Expert-Parallel MoE Inference
EEP makes wide expert-parallel MoE serving survive single-rank failures with an 11s recovery pause, 8s reintegration pause, and throughput restored to 95% of pre-fault level within 52s while staying within 4.4% of a fixed-membership baseline in steady state.
-
Test-Time Speculation
Test-Time Speculation adapts draft models online via target-model verifications to sustain high acceptance lengths during long LLM generations.
-
Dooly: Configuration-Agnostic, Redundancy-Aware Profiling for LLM Inference Simulation
Dooly reduces LLM inference profiling costs by 56.4% via configuration-agnostic taint-based labeling and selective database reuse, delivering simulation accuracy within 5% MAPE for TTFT and 8% for TPOT across 12 models.
-
SDG-MoE: Signed Debate Graph Mixture-of-Experts
SDG-MoE introduces learned signed interaction graphs and disagreement-gated deliberation among experts in MoE architectures, yielding 19.8% better validation perplexity than the strongest baseline.
-
Approximation-Free Differentiable Oblique Decision Trees
DTSemNet gives an exact, invertible neural-network encoding of hard oblique decision trees that supports direct gradient training for both classification and regression without probabilistic softening or quantized estimators.
-
MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference
MISA routes to a small subset of indexer heads via block statistics, matching full DSA performance on LongBench with 4-8x fewer heads and 3.82x speedup while recovering over 92% of selected tokens.
-
SplatWeaver: Learning to Allocate Gaussian Primitives for Generalizable Novel View Synthesis
SplatWeaver dynamically allocates Gaussian primitives via cardinality experts and pixel-level routing guided by high-frequency cues for improved generalizable novel view synthesis.
-
When Are Experts Misrouted? Counterfactual Routing Analysis in Mixture-of-Experts Language Models
Standard top-k routers in MoE language models often select suboptimal routes for difficult tokens, and updating only the final router layer raises pass@K on AIME and HMMT benchmarks across multiple models.
-
TENNOR: Trustworthy Execution for Neural Networks through Obliviousness and Retrievals
TENNOR enables efficient private training of wide neural networks in TEEs by recasting sparsification as doubly oblivious LSH retrievals and introducing MP-WTA to cut hash table memory by 50x while preserving accuracy.
-
Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation
MoLF routes updates between full fine-tuning and LoRA at the optimizer level to match or exceed the better of either static method, with an efficient LoRA-only variant outperforming prior adaptive approaches.
-
Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration
CEA assembles per-token low-rank residual updates via dense affinities over hyper-adapter-generated components to improve all-in-one image restoration on spatially non-uniform degradations.
-
Relay Buffer Independent Communication over Pooled HBM for Efficient MoE Inference on Ascend
A buffer-free MoE dispatch and combine method on Ascend hardware with pooled HBM cuts intermediate relay overhead via direct expert window access.
-
Misrouter: Exploiting Routing Mechanisms for Input-Only Attacks on Mixture-of-Experts LLMs
Misrouter enables input-only attacks on MoE LLMs by optimizing queries on open-source surrogates to route toward weakly aligned experts and transferring them to public APIs.
-
Affinity Is Not Enough: Recovering the Free Energy Principle in Mixture-of-Experts
Adding temporal memory via LIF, precision-weighted gating, and anticipatory prediction to MoE routers recovers effective expert selection at distribution transitions, with ablation confirming a super-additive beta-ant interaction.
-
RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs
RouteHijack is a routing-aware jailbreak that identifies safety-critical experts via activation contrast and optimizes suffixes to suppress them, reaching 69.3% average attack success rate on seven MoE LLMs with strong transfer to variants and VLMs.
-
MASCing: Configurable Mixture-of-Experts Behavior via Activation Steering Masks
MASCing uses an LSTM surrogate and optimized steering masks to enable flexible, inference-time control over MoE expert routing for safety objectives, improving jailbreak defense and content generation success rates substantially across multiple models.
-
Adaptive and Fine-grained Module-wise Expert Pruning for Efficient LoRA-MoE Fine-Tuning
DMEP prunes experts module-by-module in LoRA-MoE and removes load balancing after pruning, cutting trainable parameters 35-43% and raising throughput ~10% while matching or exceeding uniform baselines on reasoning tasks.
-
Hallo-Live: Real-Time Streaming Joint Audio-Video Avatar Generation with Asynchronous Dual-Stream and Human-Centric Preference Distillation
Hallo-Live achieves 20.38 FPS real-time text-to-audio-video avatar generation with 0.94s latency using asynchronous dual-stream diffusion and HP-DMD preference distillation, matching teacher model quality at 16x higher throughput.
-
Expert Upcycling: Shifting the Compute-Efficient Frontier of Mixture-of-Experts
Expert upcycling duplicates experts in an existing MoE checkpoint and continues pre-training to match fixed-size baseline performance with 32% less compute.
-
Using large language models for embodied planning introduces systematic safety risks
LLM planners for robots often produce dangerous plans even when planning succeeds, with safety awareness staying flat as model scale improves planning ability.
-
Depth Adaptive Efficient Visual Autoregressive Modeling
DepthVAR adaptively allocates per-token computational depth in VAR models using a cyclic rotated scheduler and dynamic layer masking to achieve 2.3-3.1x inference speedup with minimal quality loss.
-
Equifinality in Mixture of Experts: Routing Topology Does Not Determine Language Modeling Quality
Routing topology in sparse Mixture-of-Experts models does not determine asymptotic language modeling perplexity; multiple variants including cosine-similarity routing achieve statistically equivalent performance.
-
Sign-to-Speech Prosody Transfer via Sign Reconstruction-based GAN
SignRecGAN trains on separate sign and speech datasets via adversarial and reconstruction objectives to inject sign-derived prosody into TTS output using the S2PFormer model.
-
Plasticity-Enhanced Multi-Agent Mixture of Experts for Dynamic Objective Adaptation in UAVs-Assisted Emergency Communication Networks
PE-MAMoE combines sparsely gated mixture-of-experts actors with a non-parametric phase controller in MAPPO to maintain plasticity under dynamic user mobility and traffic, yielding 26.3% higher normalized IQM return in simulations.
-
SPAMoE: Spectrum-Aware Hybrid Operator Framework for Full-Waveform Inversion
SPAMoE reduces average MAE by 44.4% on OpenFWI datasets for full-waveform inversion via a spectral-preserving DINO encoder and dynamic frequency-band routing to specialized neural operators.
-
A Mixture of Experts Foundation Model for Scanning Electron Microscopy Image Analysis
A mixture-of-experts transformer foundation model pretrained on diverse SEM images enables generalization across materials and outperforms SOTA on unsupervised defocus-to-focus restoration.
-
SARES-DEIM: Sparse Mixture-of-Experts Meets DETR for Robust SAR Ship Detection
SARES-DEIM achieves 76.4% mAP50:95 and 93.8% mAP50 on HRSID by routing SAR features through sparse frequency and wavelet experts plus a high-resolution preservation neck, outperforming prior YOLO and SAR detectors.
-
Diffusion Policy with Bayesian Expert Selection for Active Multi-Target Tracking
A Bayesian expert selection framework with variational Bayesian last layers and lower confidence bounds improves diffusion policies for active multi-target tracking.
-
Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
A recurrent-depth architecture enables language models to improve reasoning performance by iterating computation in latent space, achieving gains equivalent to much larger models on benchmarks.
-
Scaling and evaluating sparse autoencoders
K-sparse autoencoders with dead-latent fixes produce clean scaling laws and better feature quality metrics that improve with size, shown by training a 16-million-latent model on GPT-4 activations.
-
Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
Switch Transformers use top-1 expert routing in a Mixture of Experts setup to scale to trillion-parameter language models with constant compute and up to 4x speedup over T5-XXL.
-
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
T5 casts all NLP tasks as text-to-text generation, systematically explores pre-training choices, and reaches strong performance on summarization, QA, classification and other tasks via large-scale training on the Colossal Clean Crawled Corpus.
-
Overtrained, Not Misaligned
Emergent misalignment arises from overtraining after primary task convergence and is preventable by early stopping, which retains 93% of task performance on average.
-
Domain Restriction via Multi SAE Layer Transitions
Multi-layer SAE transitions capture domain-specific signatures that distinguish OOD texts in Gemma-2 models.
-
Enabling Performant and Flexible Model-Internal Observability for LLM Inference
DMI-Lib delivers 0.4-6.8% overhead for offline batch LLM inference and ~6% for moderate online serving while exposing rich internal signals across backends, cutting latency overhead 2-15x versus prior observability baselines.
-
GenMed: A Pairwise Generative Reformulation of Medical Diagnostic Tasks
GenMed uses diffusion models to capture P(X,Y) for medical tasks and performs inference via gradient-based test-time optimization, supporting arbitrary observation combinations without retraining.
-
DuetFair: Coupling Inter- and Intra-Subgroup Robustness for Fair Medical Image Segmentation
DuetFair couples inter-subgroup adaptation with intra-subgroup robustness via FairDRO (dMoE plus subgroup-conditioned DRO) to boost worst-case and equity-scaled performance on medical segmentation benchmarks.
-
DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe Parallelism
DisagMoE achieves up to 1.8x faster MoE training by disaggregating attention and FFN layers into disjoint GPU groups with a multi-stage uni-directional pipeline and roofline-based bandwidth balancing.
-
Sparse Layers are Critical to Scaling Looped Language Models
Looped MoE models scale better than standard transformers because different experts activate on each loop pass, recovering expressivity without extra parameters, and support superior early exits.
-
XPERT: Expert Knowledge Transfer for Effective Training of Language Models
XPERT extracts and reuses cross-domain expert knowledge from pre-trained MoE LLMs via inference analysis and tensor decomposition to improve performance and convergence in downstream language model training.
-
DIMoE-Adapters: Dynamic Expert Evolution for Continual Learning in Vision-Language Models
DIMoE-Adapters uses self-calibrated expert evolution and prototype-guided selection to dynamically grow and allocate experts, outperforming prior continual learning methods on vision-language models.
-
Hierarchical Mixture-of-Experts with Two-Stage Optimization
Hi-MoE uses two-level hierarchical routing objectives to enforce group-level balance while promoting within-group specialization, yielding better perplexity and expert utilization than prior MoE baselines in NLP and vision tasks.
-
UniPool: A Globally Shared Expert Pool for Mixture-of-Experts
A shared global expert pool in MoE improves validation loss over per-layer experts and allows sublinear expert-parameter growth with depth.
-
Federation of Experts: Communication Efficient Distributed Inference for Large Language Models
FoE restructures MoE blocks into per-KV-head clusters with sum-based synchronization, removing all-to-all communication in single-node settings and limiting it to intra-node in multi-node settings for up to 5.2x faster inference with comparable quality.
-
VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading
VisMMoE exploits visual-expert affinity via token pruning to achieve up to 2.68x faster VL-MoE inference on memory-constrained hardware while keeping accuracy competitive.
-
Piper: Efficient Large-Scale MoE Training via Resource Modeling and Pipelined Hybrid Parallelism
Piper introduces resource modeling and pipelined hybrid parallelism for MoE training, delivering 2-3.5X higher MFU than prior frameworks and 1.2-9X better all-to-all bandwidth.