↓Skip to main content

🏢 Microsoft Research

How to Solve Contextual Goal-Oriented Problems with Offline Datasets?
·2005 words·10 mins· loading · loading
Machine Learning Reinforcement Learning 🏢 Microsoft Research
CODA: A novel method for solving contextual goal-oriented problems with offline datasets, using unlabeled trajectories and context-goal pairs to create a fully labeled dataset, outperforming other bas…
ElasTST: Towards Robust Varied-Horizon Forecasting with Elastic Time-Series Transformer
·2198 words·11 mins· loading · loading
Machine Learning Deep Learning 🏢 Microsoft Research
ElasTST: A novel time-series transformer enables robust forecasting across various horizons without per-horizon training, enhancing adaptability and accuracy.
EEG2Video: Towards Decoding Dynamic Visual Perception from EEG Signals
·2189 words·11 mins· loading · loading
Computer Vision Video Understanding 🏢 Microsoft Research
EEG2Video reconstructs dynamic videos from EEG signals, achieving 79.8% accuracy in semantic classification and 0.256 SSIM in video reconstruction.
Does Reasoning Emerge? Examining the Probabilities of Causation in Large Language Models
·2327 words·11 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
LLMs’ reasoning abilities are assessed via a novel framework that leverages probabilities of causation, revealing that while capable, their understanding of causality falls short of human-level reason…
Diffusion Model with Cross Attention as an Inductive Bias for Disentanglement
·3205 words·16 mins· loading · loading
Representation Learning 🏢 Microsoft Research
Diffusion models with cross-attention: a powerful inductive bias for effortless disentanglement!
DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs
·2495 words·12 mins· loading · loading
Multimodal Learning Vision-Language Models 🏢 Microsoft Research
DeepStack: Stacking visual tokens boosts LMMs efficiency and performance!
CulturePark: Boosting Cross-cultural Understanding in Large Language Models
·2738 words·13 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
CulturePark, a novel multi-agent communication framework, generates high-quality cross-cultural data to fine-tune LLMs, significantly reducing cultural bias and boosting cross-cultural understanding.
CultureLLM: Incorporating Cultural Differences into Large Language Models
·2507 words·12 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
CultureLLM, a new approach, effectively incorporates cultural nuances into LLMs using semantic data augmentation, significantly outperforming existing models.
Compositional 3D-aware Video Generation with LLM Director
·2894 words·14 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
LLM-directed compositional 3D-aware video generation (C3V) achieves high-fidelity video generation with diverse motion and flexible concept control by decomposing prompts, generating 3D concepts, and …
Can large language models explore in-context?
·4498 words·22 mins· loading · loading
AI Generated Natural Language Processing Large Language Models 🏢 Microsoft Research
LLMs struggle with in-context exploration, needing substantial prompt engineering or training interventions to effectively explore multi-armed bandit environments.
Boosting Text-to-Video Generative Model with MLLMs Feedback
·2610 words·13 mins· loading · loading
Multimodal Learning Vision-Language Models 🏢 Microsoft Research
MLLMs enhance text-to-video generation by providing 135k fine-grained video preferences, creating VIDEOPREFER, and a novel reward model, VIDEORM, boosting video quality and alignment.
ALPINE: Unveiling The Planning Capability of Autoregressive Learning in Language Models
·2122 words·10 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
ALPINE reveals how Transformer-based LLMs learn planning by embedding graph information into their weights, but also highlights their inability to handle transitive relationships.
Aligning LLM Agents by Learning Latent Preference from User Edits
·2688 words·13 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
PRELUDE, a novel framework, leverages user edits of LLM outputs to learn latent preferences, improving agent alignment and minimizing edit costs. CIPHER, its efficient algorithm, infers preferences f…
Advancing Spiking Neural Networks for Sequential Modeling with Central Pattern Generators
·1882 words·9 mins· loading · loading
🏢 Microsoft Research
Bio-inspired CPG-PE enhances spiking neural networks’ sequential modeling by efficiently encoding position information, outperforming conventional methods across various tasks.
Adam with model exponential moving average is effective for nonconvex optimization
·281 words·2 mins· loading · loading
AI Theory Optimization 🏢 Microsoft Research
Clipped Adam with EMA achieves optimal convergence rates for smooth and non-smooth nonconvex optimization, particularly when scales vary across different coordinates.
Accuracy is Not All You Need
·5583 words·27 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Microsoft Research
LLM compression accuracy hides crucial behavioral changes; use % flips and KL-divergence for better evaluation.
A Study of Plasticity Loss in On-Policy Deep Reinforcement Learning
·2749 words·13 mins· loading · loading
Reinforcement Learning 🏢 Microsoft Research
On-policy deep RL agents suffer from plasticity loss, but this paper introduces ‘regenerative’ methods that consistently mitigate this, improving performance in challenging environments.
A Bayesian Approach to Data Point Selection
·3079 words·15 mins· loading · loading
AI Generated Machine Learning Deep Learning 🏢 Microsoft Research
BADS: a novel Bayesian approach to data point selection efficiently optimizes deep learning models by jointly inferring instance weights and model parameters using stochastic gradient Langevin dynamic…