catastrophic forgetting in llms, continual learning
Severe forgetting during training.
3,145 technical terms and definitions
Severe forgetting during training.
When fine-tuning on new data causes the model to forget previously learned knowledge.
Category management groups similar materials for coordinated sourcing strategies and supplier relationships.
Heavy-tailed robust loss.
Standard left-to-right prediction.
Mask preventing attention to future tokens in autoregressive models.
Track causal paths through network.
Causality-Aware Walk aggregates information from temporal neighborhoods respecting causal constraints in dynamic graphs.
Convolutional Block Attention Module combines channel and spatial attention sequentially.
Convergent Cross Mapping infers causality in dynamical systems by testing whether manifold reconstructions from one variable can predict another.
Smooth version of ELU.
Measure representation similarity.
Certified fairness provides mathematical guarantees of equitable treatment.
Formal guarantees of robustness.
Formal guarantees that model predictions won't change within certain input perturbations.
Chain-of-thought prompting asks model to show reasoning steps. Improves accuracy on math, logic, and complex tasks.
Train models to reason step-by-step.
Encourage step-by-step reasoning.
Ask model to show reasoning steps before giving final answer.
Visual reasoning with explicit steps.
Chain-of-thought prompting requests step-by-step reasoning improving complex problem solving.
Chainlit builds chat interfaces for LLM apps. Streaming, memory, tools. Production-ready.
Change point detection identifies times when statistical properties of time series change indicating regime shifts.
Channel attention mechanisms adaptively weight feature channels based on global context.
Channel shuffle permutes channels between groups enabling information flow in grouped convolutions.
Charge-induced voltage alteration uses localized charge injection to modulate circuit behavior localizing sensitive nodes.
Device itself charged then discharged.
Design ESD protection for CDM.
Chat models are optimized for multi-turn conversational interactions.
Conversational AI based on GPT models with RLHF.
Spectral GNN using Chebyshev polynomials.
ChebNet approximates spectral graph convolutions using Chebyshev polynomials reducing computational cost.
Distribute checkpoint across nodes.
Saved snapshot of model weights during training.
Checkpoints save model weights periodically. Resume training from checkpoint if interrupted. Keep best checkpoint by validation loss.
Chemical decapsulation dissolves epoxy molding compound using hot acids preserving die integrity.
Identify chemical compounds in text.
Chemical recycling recovers and purifies process chemicals for reuse reducing consumption and waste.
Chemical waste from semiconductor processing requires specialized handling neutralization and disposal due to hazardous properties.
Named entity recognition for chemistry.
Chilled water optimization adjusts setpoints and flow to minimize chiller and pump energy.
Models trained optimally.
Optimal data-to-parameter ratio.
Optimal compute allocation: scale model and data equally.
DeepMind's efficiently trained language model.
Circuit breakers halt requests to failing services allowing recovery time.
Find functional circuits in networks.
Circular economy approaches maximize material reuse refurbishment and recycling minimizing waste in semiconductor manufacturing lifecycle.
Analyze legal citation networks.
Identify factual claims in text.