↓Skip to main content

Posters

2024

3D Equivariant Pose Regression via Direct Wigner-D Harmonics Prediction
·2707 words·13 mins· loading · loading
Computer Vision 3D Vision 🏢 Pohang University of Science and Technology
3D pose estimation is revolutionized by a novel SO(3)-equivariant network directly predicting Wigner-D harmonics, achieving state-of-the-art accuracy and efficiency.
3-in-1: 2D Rotary Adaptation for Efficient Finetuning, Efficient Batching and Composability
·2315 words·11 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Language Technology Lab, University of Amsterdam
RoAd: a novel parameter-efficient finetuning method uses 2D rotation to adapt LLMs, enabling efficient batching, composability, and improved interpretability.
2DQuant: Low-bit Post-Training Quantization for Image Super-Resolution
·2009 words·10 mins· loading · loading
AI Generated Computer Vision Image Generation 🏢 Shanghai Jiao Tong University
2DQuant achieves highly efficient and accurate low-bit image super-resolution by using a dual-stage post-training quantization method that minimizes accuracy loss in transformer-based models, surpassi…
2D-OOB: Attributing Data Contribution Through Joint Valuation Framework
·2147 words·11 mins· loading · loading
AI Theory Interpretability 🏢 University of Illinois Urbana-Champaign
2D-OOB: a novel framework for jointly attributing data values to individual features, enabling fine-grained outlier detection and improved model performance.
$SE(3)$ Equivariant Ray Embeddings for Implicit Multi-View Depth Estimation
·2436 words·12 mins· loading · loading
Computer Vision 3D Vision 🏢 Toyota Research Institute
SE(3)-equivariant ray embeddings in Perceiver IO achieve state-of-the-art implicit multi-view depth estimation, surpassing methods that rely on data augmentation for approximate equivariance.
$psilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise
·1776 words·9 mins· loading · loading
Machine Learning Deep Learning 🏢 Faculty of Computing, Harbin Institute of Technology
e-Softmax: A simple plug-and-play module enhances deep learning model robustness against noisy labels by approximating one-hot vectors, achieving noise-tolerant learning with controllable excess risk.
$eta$-DPO: Direct Preference Optimization with Dynamic $eta$
·2106 words·10 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Alibaba Group
β-DPO dynamically adjusts a key parameter in Direct Preference Optimization, significantly improving LLM alignment with human preferences.
$C^2M^3$: Cycle-Consistent Multi-Model Merging
·3768 words·18 mins· loading · loading
Machine Learning Federated Learning 🏢 Sapienza University of Rome
C2M³: A novel data-free method ensures cycle-consistent merging of neural networks, significantly improving model aggregation across various architectures and datasets.
$ extit{Trans-LoRA}$: towards data-free Transferable Parameter Efficient Finetuning
·3529 words·17 mins· loading · loading
AI Generated Natural Language Processing Large Language Models 🏢 MIT-IBM Watson AI Lab
Trans-LoRA enables near data-free transfer of fine-tuned LLMs across models!
$ extit{Read-ME}$: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Design
·2049 words·10 mins· loading · loading
Natural Language Processing Large Language Models 🏢 University of Texas at Austin
Read-ME refactors pre-trained dense LLMs into efficient, router-decoupled Mixture-of-Experts (MoEs) via activation sparsity, achieving up to 10.1% improvement on MMLU and 6.1% reduction in latency.
$ extit{NeuroPath}$: A Neural Pathway Transformer for Joining the Dots of Human Connectomes
·2210 words·11 mins· loading · loading
AI Applications Healthcare 🏢 University of North Carolina at Chapel Hill
NeuroPath: A novel deep learning model reveals how brain structure supports brain function by uncovering multi-hop neural pathways, improving brain network analysis accuracy.
$ extit{Bifr"ost}$: 3D-Aware Image Compositing with Language Instructions
·3407 words·16 mins· loading · loading
Multimodal Learning Vision-Language Models 🏢 Hong Kong University of Science and Technology
Bifröst: A novel 3D-aware framework for instruction-based image compositing, leveraging depth maps and an MLLM for high-fidelity results.
$ ext{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition
·1939 words·10 mins· loading · loading
Computer Vision Face Recognition 🏢 Tencent Youtu Lab
ID³: A novel diffusion model generates diverse, identity-preserving synthetic face datasets for accurate and privacy-preserving face recognition, exceeding current state-of-the-art methods.
$ ext{Di}^2 ext{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation
·2529 words·12 mins· loading · loading
AI Generated Computer Vision 3D Vision 🏢 Hong Kong University of Science and Technology
Di²Pose, a novel discrete diffusion model, tackles occluded 3D human pose estimation by employing a two-stage process: pose quantization and discrete diffusion, achieving state-of-the-art results.
(FL)$^2$: Overcoming Few Labels in Federated Semi-Supervised Learning
·2049 words·10 mins· loading · loading
AI Generated Machine Learning Federated Learning 🏢 KAIST
Federated Semi-Supervised Learning (FSSL) struggles with limited labeled data. (FL)² bridges this gap using adaptive thresholding, sharpness-aware consistency regularization, and learning status-awar…