Open-Source Software and Frameworks

Caesar is an autonomous AI research agent designed to move beyond retrieval toward deeper exploration, creative synthesis, and discovery. 

TerraLingua is a multi-agent simulation framework for studying emergent behavior, artifact creation, and cultural evolution.

Quantized Evolution Strategies (QES) is an optimization paradigm that performs full-parameter fine-tuning directly in the quantized space. 

Evolution Strategy for Metacognitive Alignment (ESMA) is a novel method to improve large language models’ awareness of their own knowledge.

Evolution strategies (ES) is used to directly optimize billions of parameters of large language models (LLMs).

neuro-san is data-driven multi-agent orchestration framework designed to simplify and accelerate the development of collaborative AI systems.

Realizing Human Expertise through AI (RHEA) enhances diverse human-developed solutions by evolving them through AI.

Semantic density measures uncertainty in LLM responses and works out of the box for free-form generation tasks.

Language Model Crossover (LMX) uses LLMs as the engine of evolution, driving recombination and variation for any task where solutions are representable as text.

Shortest Edit Path Crossover (SEPX) solves the permutation problem in neural architecture search by recombining architectures in graph space.

AQuaSurF uses a surrogate modeling approach to quickly discover new activation functions that improve performance on a variety of tasks.

Act-Bench datasets provide training results for 2,913 activation functions, enabling quick benchmarking of new AutoML algorithms.

AutoInit optimizes deep learning initialization for robust learning and integrates with TensorFlow experiments.

TaylorGLO evolves loss functions with multivariate Taylor polynomials for automatic regularization.

The Traveling Observer Model (TOM) implements deep multi-task learning through spatial variable embeddings.

Residual-based error detection (RED) extends RIO to classification, identifying misclassifications, out-of-distribution, and adversarial inputs.

This package provides predictors, prescriptors, and evaluation code for forecasting COVID-19 cases and interventions.

Residual estimation with input/output kernels (RIO) quantifies confidence and enhances point-prediction accuracy.

Modular Universal Reparameterization (MUiR) enables multitask learning across language, vision, and genetics.