✨LFM2 Technical Report

📝 Summary:
LFM2 is a family of compact foundation models designed for efficient on-device deployment. It uses hardware-in-the-loop architecture search and advanced training to achieve high performance across diverse tasks, including multimodal applications.

🔹 Publication Date: Published on Nov 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.23404
• PDF: https://arxiv.org/pdf/2511.23404

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#FoundationModels #EdgeAI #MultimodalAI #AIResearch #MachineLearning

✨UniQL: Unified Quantization and Low-rank Compression for Adaptive Edge LLMs

📝 Summary:
UniQL unifies quantization and low-rank compression to deploy LLMs on mobile devices. It reduces memory by 4x-5.7x and improves token throughput by 2.7x-3.4x, maintaining accuracy across various model types.

🔹 Publication Date: Published on Dec 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03383
• PDF: https://arxiv.org/pdf/2512.03383
• Project Page: https://hychiang.info/projects/uniql/
• Github: https://github.com/enyac-group/UniQL

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#LLMs #EdgeAI #Quantization #ModelCompression #DeepLearning

✨AutoNeural: Co-Designing Vision-Language Models for NPU Inference

📝 Summary:
AutoNeural is an NPU-native VLM co-designed for efficient edge inference. It uses a MobileNetV5-style vision backbone for stable integer quantization and a hybrid SSM-Transformer language backbone. This design reduces quantization errors and latency, improving real-time performance on edge devices.

🔹 Publication Date: Published on Dec 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.02924
• PDF: https://arxiv.org/pdf/2512.02924

🔹 Models citing this paper:
• https://huggingface.co/NexaAI/AutoNeural

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#AutoNeural #VisionLanguageModels #EdgeAI #AIHardware #EfficientAI

✨Real-Time Object Detection Meets DINOv3

📝 Summary:
DEIMv2 extends DEIM with DINOv3 features, achieving superior real-time object detection across GPU, edge, and mobile. It uses a Spatial Tuning Adapter and pruned HGNetv2 for diverse models, setting new state of the art with impressive performance-cost trade-offs.

🔹 Publication Date: Published on Sep 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2509.20787
• PDF: https://arxiv.org/pdf/2509.20787
• Project Page: https://intellindust-ai-lab.github.io/projects/DEIMv2/
• Github: https://github.com/Intellindust-AI-Lab/DEIMv2

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ObjectDetection #RealTimeAI #ComputerVision #MachineLearning #EdgeAI

✨HyperVL: An Efficient and Dynamic Multimodal Large Language Model for Edge Devices

📝 Summary:
HyperVL is an efficient multimodal large language model for edge devices. It uses image tiling, a Visual Resolution Compressor, and Dual Consistency Learning to reduce memory, latency, and power. HyperVL maintains performance, making it practical for on-device inference.

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14052
• PDF: https://arxiv.org/pdf/2512.14052

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#HyperVL #MLLM #EdgeAI #EfficientAI #OnDeviceAI

✨Bitnet.cpp: Efficient Edge Inference for Ternary LLMs

📝 Summary:
Bitnet.cpp enhances edge inference for ternary LLMs using a novel mixed-precision matrix multiplication library. This system incorporates Ternary Lookup Tables and Int2 with a Scale for efficient, lossless inference, achieving up to a 6.25x speed increase over baselines.

🔹 Publication Date: Published on Feb 17, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2502.11880
• PDF: https://arxiv.org/pdf/2502.11880
• Github: https://github.com/microsoft/BitNet/tree/paper

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#LLM #EdgeAI #MachineLearning #DeepLearning #AI

✨Flavors of Moonshine: Tiny Specialized ASR Models for Edge Devices

📝 Summary:
Flavors of Moonshine are tiny monolingual ASR models for underrepresented languages. They outperform larger multilingual models by using balanced data, achieving 48% lower error rates. This enables accurate on-device speech recognition.

🔹 Publication Date: Published on Sep 2, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2509.02523
• PDF: https://arxiv.org/pdf/2509.02523
• Github: https://github.com/moonshine-ai/moonshine

🔹 Models citing this paper:
• https://huggingface.co/UsefulSensors/moonshine-tiny-ja
• https://huggingface.co/UsefulSensors/moonshine-tiny-ar
• https://huggingface.co/UsefulSensors/moonshine-tiny-zh

✨ Spaces citing this paper:
• https://huggingface.co/spaces/wmoto-ai/moonshine-tiny-ja-demo

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ASR #EdgeAI #LowResourceLanguages #MachineLearning #TinyML

✨MNN: A Universal and Efficient Inference Engine

📝 Summary:
MNN is an efficient deep learning inference engine for mobile devices. It addresses compatibility and resource limits through pre-inference, kernel optimization, and backend abstraction, outperforming other lightweight frameworks.

🔹 Publication Date: Published on Feb 27, 2020

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2002.12418
• PDF: https://arxiv.org/pdf/2002.12418
• Github: https://github.com/alibaba/MNN

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#DeepLearning #MobileAI #EdgeAI #Optimization #MachineLearning

✨ECoLAD: Deployment-Oriented Evaluation for Automotive Time-Series Anomaly Detection

📝 Summary:
ECoLaD is a new framework evaluating time-series anomaly detection under compute constraints, critical for in-vehicle systems. It uses efficiency reductions to assess feasibility. Findings show classical methods sustain performance, but deep learning often becomes infeasible before losing accuracy.

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10926
• PDF: https://arxiv.org/pdf/2603.10926

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#AnomalyDetection #TimeSeries #AutomotiveAI #EdgeAI #DeepLearning

✨Efficient Universal Perception Encoder

📝 Summary:
EUPE enhances edge device performance through a novel two-stage knowledge distillation approach. It scales up to a large proxy teacher then down to an efficient encoder. This method provides superior, versatile representations for diverse tasks, outperforming prior techniques.

🔹 Publication Date: Published on Mar 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.22387
• PDF: https://arxiv.org/pdf/2603.22387
• Github: https://github.com/facebookresearch/eupe

🔹 Models citing this paper:
• https://huggingface.co/facebook/EUPE-ConvNeXt-S
• https://huggingface.co/facebook/EUPE-ViT-S
• https://huggingface.co/facebook/EUPE-ViT-B

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#KnowledgeDistillation #EdgeAI #ComputerVision #DeepLearning #RepresentationLearning

🔥 Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on Heterogeneous Robots

💡 The paper introduces Embodied.cpp, a portable C++ runtime that enables efficient deployment of embodied AI models across heterogeneous edge devices. The problem addressed is the fragmentation of embodied AI model deployment, which is currently limited by model-specific Python stacks, backend assumptions, and robot-side glue code. This makes it difficult to deploy these models on various devices, especially on heterogeneous edge devices.

The authors propose Embodied.cpp as a solution, which is based on an architectural analysis of representative vision-language-action and world-action models. The runtime is organized into five layers: input adapters, sequence builders, backbone execution, head plugins, and deployment adapters. This modular design provides multi-rate execution, latency-first fused inference, and extensible operator and I/O support, allowing for deployment across diverse devices, robots, and simulators through a single backend abstraction.

The results show that Embodied.cpp achieves successful closed-loop execution with high task success rates on two vision-language-action models, and reduces block memory usage on a preliminary world-action model benchmark. Specifically, the VLA deployments achieve 100.0% and 91.0% task success rates, while the WAM benchmark reduces block memory from 312.2 MiB to 88.1 MiB. These results demonstrate that Embodied.cpp improves deployment efficiency while preserving high accuracy across diverse embodied model architectures. Overall, the paper contributes a portable and efficient runtime for embodied AI models, enabling their deployment on a wide range of devices and robots.

📅 Published on Jul 2

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2607.02501
• PDF: https://arxiv.org/pdf/2607.02501

🤖 Models citing this paper:
• https://huggingface.co/SEU-PAISys/Embodied.cpp

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#EmbodiedAI #HeterogeneousRobots #EdgeAI #RoboticsEngineering #AIModelDeployment