Architecture Design (Transformers, SSMs, MoE)

Dynamic layer routing can boost LLM accuracy by 5% without the need for weight updates or expensive search loops.

Geometry-adaptive prompting can transform how we approach few-shot learning in dynamic graphs, leading to substantial performance gains.

MicroEvo achieves a staggering 10.6x increase in search efficiency while improving Pareto-front quality by 36.2%, revolutionizing microarchitecture design exploration.

Dense-Cast achieves a remarkable MAE of 0.235 mm for half-hourly precipitation nowcasting, setting a new benchmark for accuracy in this challenging domain.

A runtime observability framework reveals how modern memory architectures can be rigorously monitored and quantified, exposing hidden risks in AI model performance.

A single model can now adaptively balance short-term precision and long-term accuracy in weather forecasts, revolutionizing how we approach atmospheric predictions.

SiPE establishes a new frontier in Transformer architecture by effectively integrating syntactic structure, leading to substantial gains in both syntax and general language understanding without compromising efficiency.

JTA reveals that validation gaps in safety-critical software can be systematically addressed by treating scenarios, test systems, and systems under test as interconnected entities.

U-Nets unexpectedly show greater robustness to resolution changes than anticipated, challenging assumptions about neural operator architectures in inverse imaging.

ShimGen not only matches but surpasses manually-designed protocols in consistency, revealing critical performance gains in heterogeneous memory systems.

Achieving a staggering 7,300x reduction in instruction overhead for sensor reads could redefine efficiency benchmarks in embedded systems.

ASTELD uncovers a critical gap in the autonomous AI landscape: no evaluated systems achieve both local-first deployment and enterprise-grade security.

Regime-aware modeling with MGSB boosts leak detection performance by over 20% in out-of-distribution scenarios compared to traditional methods.

Recurrent Vision Transformers can outperform standard models in accuracy-to-parameter trade-offs when memory constraints are prioritized, challenging conventional wisdom about architectural efficiency.

Novel aggregation techniques for Memory Augmented Autoencoders in Federated Learning can enhance anomaly detection performance, even in shallow models.

Pruning Echo State Networks dynamically can enhance forecasting accuracy while significantly reducing model complexity.

Elbow-based routing can cut inference latency by over 5% in MoE models while preserving accuracy, revolutionizing expert selection efficiency.

Merging experts based on their phase roles can enhance MoE-VLM performance by up to 9.6%, challenging the effectiveness of traditional global aggregation methods.

MESH achieves a 62.5% reduction in optimizer-state memory for Mixture-of-Experts training while preserving performance, challenging assumptions about memory efficiency in deep learning.

Achieving a 3.56× increase in decoding throughput without sacrificing accuracy, BinaryPC revolutionizes efficiency in long-context LLMs.

ORACLE achieves up to 104.4x faster circuit design while meeting nearly all target specifications, revolutionizing multi-objective optimization in analog circuit design.

Eigenius not only validates scientific conclusions but also uncovers discrepancies in published research, revolutionizing how we ensure data integrity in AI-driven science.

Current LLMs fail to meet the rigorous demands of PCB routing, showing major weaknesses in path planning and constraint adherence.

MCHA achieves unprecedented performance speedups for parallel-sequential computing tasks, outperforming NVIDIA A100 GPUs by up to 2456.96×.

K-EXAONE 2.0 achieves over three times the capacity of its predecessor while enhancing multilingual capabilities and context understanding.

EdgeXpert slashes LLM inference latency by over 56% while cutting energy use by nearly 45%, all without sacrificing accuracy.

Achieving up to 91.4x speedup in O-RAN fronthaul decompression could revolutionize the efficiency of 5G networks.

AFD-Ledger reveals that optimizing deployment for AFD can drastically cut evaluation costs while exposing the nuanced performance dynamics between homogeneous and heterogeneous setups.

PowerScope achieves intra-cycle power estimation with 80x speedup and competitive accuracy, revolutionizing power analysis workflows.

Achieving up to 71% lower error in activation functions while using less area and power could revolutionize the efficiency of neural network accelerators.

Fragmented agentic AI workflows expose significant inefficiencies in conventional server architectures, necessitating a radical rethink of resource allocation strategies.

Formal verification of a compiler for asynchronous dataflow could redefine the reliability and efficiency of parallel computing architectures.

Muon optimizer's unique advantage lies in its ability to enhance token efficiency specifically when applied to the output projection, challenging assumptions about conditioning in state-space models.

LAEF achieves superior point-of-care ECG diagnostics by leveraging lead-agnostic architecture, outperforming traditional models even with minimal lead data.

Energy-aware DNN design can now be optimized offline, paving the way for truly autonomous AI on intermittent power sources.

Dynamic processing in Transformers can exceed static processing, revealing a deeper layer of interaction that mirrors human language processing.

Tiny input changes can destabilize UAV tracking models, revealing a new attack surface that undermines their efficiency and accuracy.

MuEvo not only evolves heuristic ensembles but also dynamically adapts component priorities, leading to superior performance in complex optimization tasks.

AgenticECO clears 7 out of 9 defect cases with minimal disturbance, revolutionizing ECO processes in 3D-ICs by ensuring repair attribution without the chaos of full rerouting.

LoopMTP boosts reasoning accuracy by up to 8.1% by effectively guiding looped transformer iterations with multi-token prediction.

MoEGen achieves instance-specific adaptations without the storage burden of full LoRA experts, revolutionizing how we think about parameter-efficient fine-tuning.

AcceptMoE slashes host-to-device traffic by over 73% while boosting throughput by more than double, all without a major accuracy trade-off.

Achieving competitive accuracy while slashing computational costs, this method redefines the feasibility of 3D object detection on lightweight devices.

SLAMFormer-$\infty$ can handle trajectory sequences over 17 km without losing performance, revolutionizing long-range SLAM capabilities.

Real-time AI monitoring can boost tomato crop disease detection rates to 95%, transforming agriculture scalability.

Fovea achieves a remarkable 7.80x speedup in wafer architecture selection while ensuring optimal performance across diverse workloads.

CAMTA achieves nearly an order of magnitude improvement in Softmax function approximation while offering unprecedented runtime configurability in hardware.

ALiBi positional encoding can blind attention heads, drastically impairing token retrieval without impacting standard performance metrics.

MoE dLLMs can outperform leading models with significantly fewer training tokens, challenging assumptions about data efficiency in large-scale language models.

Recovering multi-head attention parameters without orthogonality assumptions could revolutionize how we learn complex attention mechanisms in neural networks.