#computational-efficiency

[ follow ]
#manifold-constrained-hyper-connections-mhc #model-scaling #training-stability #mxfp4 #data-types #ai-models
Artificial intelligence
fromBusiness Insider
2 months ago

China's DeepSeek kicked off 2026 with a new AI training method that analysts say is a 'breakthrough' for scaling

DeepSeek developed Manifold-Constrained Hyper-Connections (mHC), a training method that enables richer internal model communication while preserving training stability and efficiency as models scale.
fromTheregister
7 months ago

OpenAI gpt-oss LLMs use MXFP4: smaller, faster, cheaper

MXFP4 is a 4-bit floating point data type defined by the Open Compute Project, allowing massive compute savings compared to traditional data types used by LLMs.
Artificial intelligence
Artificial intelligence
fromHackernoon
1 year ago

Multi-Token Prediction: Mastering Algorithmic Reasoning with Enhanced Resource Use | HackerNoon

Multi-token prediction in training language models allows for efficient resource allocation based on token prediction difficulty.
Data science
fromHackernoon
1 year ago

Battle of the Algorithms: Why SGRLD Beats the Competition in GP Inference | HackerNoon

The SGRLD method significantly improves the estimation of spatial covariance parameters in large datasets compared to traditional Bayesian methods.
Artificial intelligence
fromWIRED
10 months ago

Google DeepMind's AI Agent Dreams Up Algorithms Beyond Human Expertise

AlphaEvolve demonstrates that AI models can generate novel and efficient algorithms that surpass human capabilities in specific tasks.
[ Load more ]