29,259 technical terms and definitions
adamw, l2 regularization, decoupled weight decay, optimizer regularization
Weight decay shrinks trainable parameters during optimization to limit uncontrolled weight growth and improve generalization. It is a core regularizer in vision and language…
l2 regularization, decoupled weight decay, adamw weight decay, regularization strength
Weight Decay is the regularization technique that penalizes large weight values by adding a fraction of the weight magnitude to the loss function or directly shrinking weights…
computer vision
Weight Decay in Vision Transformers is a critically important, aggressively calibrated regularization hyperparameter that applies continuous, mathematically enforced shrinkage…
weight decay, decay l2, cifar l2 sgd, weight decay & l2 regularization, weight decay l2, decay l2 regularization, l2 regularization
Weight Decay & L2 Regularization Introduction & Motivation Weight decay: penalize large weights during training. Regularize model complexity. Applications: improve generalization…
weight decay l2 regularization and model complexity control, weight decay l2, decay l2 regularization, l2 regularization model, regularization model complexity, weight decay
Weight Decay L2 Regularization and Model Complexity Control Introduction & Motivation Weight Decay L2 Regularization and Model Complexity Control addresses a central problem in L2…
Weight Decay and L2 Regularization Introduction & Motivation Weight Decay and L2 Regularization addresses a central problem in optimization algorithm tuning and improvement: how…
neural architecture
Weight Entanglement is a phenomenon in weight-sharing NAS methods where the shared weights of sub-networks interfere with each other — preventing accurate performance estimation…
neural architecture search
Weight Inheritance is reusing previously trained weights when evaluating mutated or expanded architectures. - It reduces search cost by avoiding full retraining from random…
initialization, xavier initialization, glorot initialization, he initialization, kaiming initialization, weight init, parameter initialization
Weight initialization is the choice of what values a network's parameters hold *before* the first gradient step — and it is far less innocent than it sounds. Set the initial…
Xavier initialization, He initialization, muP maximal update parametrization
Weight Initialization Strategies (Xavier, He, μP) are methods for setting initial neural network weights to enable stable training with appropriate gradient flow — Xavier…
weight initialization, initialization strategies, 2l bn cifar elu gelu gpt
Weight Initialization Strategies Introduction & Motivation Weight Initialization Strategies dictate how trainable parameters in deep neural networks are assigned initial numerical…
weight initialization strategies, weight initialization, initialization strategies
Weight Initialization Strategies Introduction & Motivation Weight Initialization Strategies addresses a central problem in ml-training process control…
optimization
Weight Normalization is a reparameterization technique that decouples the magnitude and direction of weight vectors — representing each weight vector as w = g · v/ v , where g is…
quantization aware training, qat, fake quantize, ste quantization
Quantization-Aware Training (QAT) is the training technique that simulates the effects of low-bit quantization during the forward pass while maintaining full-precision gradients…
gptq quantization, awq quantization, int4 quantization, post training quantization llm
Weight Quantization for LLMs is the model compression technique that reduces the numerical precision of neural network weights from 16-bit floating point to 4-bit or 8-bit…
quantization schemes neural networks, symmetric asymmetric quantization, per channel quantization, quantization calibration
Weight Quantization Methods are the precision reduction techniques that map high-precision floating-point weights to low-bitwidth integer or fixed-point representations — using…
model optimization
Weight sharing uses the same parameters across multiple parts of a model, reducing parameter count significantly. Applications: Tied embeddings: Input and output embeddings share…
model optimization
Weight Sharing is a parameter-efficiency technique where multiple connections or structures reuse the same weights - It reduces model size and can improve regularization through…
Weight Sharing Compression Introduction & Motivation Weight Sharing Compression addresses a central problem in reducing the size and cost of compression models: how to Reduce…
Weight Sharing Compression (Extended Study) Introduction & Motivation Weight Sharing Compression (Extended Study) addresses a central problem in reducing the size and cost…
Weight Sharing Compression (Extended Study) Introduction & Motivation Weight Sharing Compression (Extended Study) addresses a central problem in reducing the size and cost…
neural architecture
Weight-Sharing Networks are neural architectures where the same set of parameters is reused across multiple computational operations — encoding the inductive bias that the same…
Weight-Sharing Supernet Search Introduction & Motivation Weight-Sharing Supernet Search addresses a central problem in automated design of nas architectures: how to Automatically…
Weight-Sharing Supernet Search (Extended Study) Introduction & Motivation Weight-Sharing Supernet Search (Extended Study) addresses a central problem in automated design of nas…
Weight-Sharing Supernet Search (Extended Study) Introduction & Motivation Weight-Sharing Supernet Search (Extended Study) addresses a central problem in automated design of nas…
weight-stationary vs output-stationary dataflow trade-offs, weight-stationary vs output-stationary, vs output-stationary dataflow, output-stationary dataflow trade-offs
Weight-Stationary vs Output-Stationary Dataflow Trade-offs Introduction & Motivation Weight-Stationary vs Output-Stationary Dataflow Trade-offs addresses a central problem…
weight update frequency and accumulation buffer trade-offs, weight update frequency, update frequency accumulation, frequency accumulation buffer, accumulation buffer trade-offs
Weight Update Frequency and Accumulation Buffer Trade-Offs Introduction & Motivation Weight Update Frequency and Accumulation Buffer Trade-Offs addresses a central problem…
weighted average, weighted arithmetic mean, weights, grade point average, GPA, weighted sum, importance weighting
A weighted mean is an average in which each value is multiplied by a weight reflecting its importance or frequency before the sum is taken, and the total is then divided by…
weighted round-robin balancing, weighted round-robin, round-robin balancing
Weighted Round-Robin Balancing Introduction & Motivation Weighted Round-Robin Balancing addresses a central problem in distributed-systems process control…
neural network weights, weight initialization
Weights are the numbers that multiply a neuron's inputs before they're summed — alongside bias terms, they are the actual learned content of a trained model, the part that starts…
weights and biases platform, wandb experiment tracking, weights biases experiment tracking, wandb machine learning platform
Weights & Biases (W&B) is a machine-learning experiment management platform used to record runs, configurations, metrics, system observations, datasets, model artifacts…
mlops
Weights & Biases is the experiment tracking and visualization platform focused on real-time monitoring and collaborative ML development - it offers rich run analytics, system…
graph neural networks
Weisfeiler-Lehman is an iterative color-refinement procedure used to characterize graph structure and bound GNN discrimination power - It repeatedly relabels nodes based on…
graph algorithms
Weisfeiler-Lehman (WL) Kernel is a graph similarity measure based on the iterative Weisfeiler-Lehman color refinement procedure — which assigns increasingly fine-grained labels…
formation, retrograde, well, process, substrate, bias
Well Formation and Retrograde Well Process is the creation of localized doped regions (wells) in semiconductor substrate enabling isolated NMOS and PMOS device regions — using…
process integration
Well Engineering is the design and formation of substrate wells to control transistor isolation, body bias, and leakage - It sets foundational electrostatic conditions that…
twin well, triple well, nwell pwell
Well Formation — creating doped regions (wells) in the silicon substrate to house NMOS and PMOS transistors, establishing the fundamental structure for CMOS circuits. Why Wells?…
well implant energy calibration drift prediction machine learning, well implant, energy calibration drift, sims, well implant energy calibration drift prediction
Well Implant Energy Calibration Drift Prediction with Machine Learning Introduction & Motivation Well Implant Energy Calibration Drift Prediction with Machine Learning addresses…
well implant profile retrograde doping prediction, well implant profile, implant profile retrograde, profile retrograde doping, retrograde doping prediction, well implant
Well Implant Profile Retrograde Doping Prediction with Machine Learning Introduction & Motivation Well Implant Profile Retrograde Doping Prediction with Machine Learning addresses…
process integration
Well implantation is dopant implantation steps that form p-well and n-well regions for transistor threshold and body control - Energy dose and anneal conditions set junction depth…
wpe, lateral channel doping, vth variation layout, well implant scatter, layout dependent effect
Well Proximity Effect (WPE) and Layout-Dependent Effects are the transistor parameter variations caused by the proximity of a device to the well or other layout features — where…
well proximity effect, wpe, design, threshold voltage mismatch, analog layout, current mirror mismatch, SRAM bit cell, dummy device, implant scattering
Well proximity effect (WPE) is a layout-dependent shift in transistor threshold voltage that depends purely on how close a transistor sits to the edge of its well — not on…
layout dependent effect, STI stress effect, transistor neighborhood effect
Well Proximity Effect (WPE) and Layout-Dependent Effects (LDE) are the systematic variations in transistor characteristics caused by the local layout context surrounding each…
well edge proximity, layout dependent effect, transistor proximity effect, systematic variation layout
Well Proximity Effect (WPE) is the layout-dependent transistor variability phenomenon where the threshold voltage and drive current of a MOSFET shift measurably depending on its…
machine learning
Welsch Loss is a robust loss function that bounds the maximum penalty for outliers — using an exponential form L(r) = (c^2)/(2)[1 - exp(-(r/c)^2)] that asymptotes to a constant…
Western Electric Company, Hawthorne Works, Hawthorne studies, Allentown transistor plant, AT&T manufacturing arm, 1984 divestiture
Western Electric was the manufacturing and supply arm of the Bell System, the company that built the telephones, switches, cables and, from the early 1950s, the transistors that…
spc
Western Electric rules is the classical SPC pattern-detection ruleset used to identify non-random behavior on control charts beyond simple limit exceedance - it improves…
koh etching, tmah etch
Wet Anisotropic Etching uses orientation-dependent etch rates in crystalline materials to create precisely shaped structures, commonly using KOH or TMAH on silicon. What Is Wet…
manufacturing equipment
Wet Bench is semiconductor process station for wafer cleaning, etching, and rinsing using controlled liquid chemistries - It is a core method in modern semiconductor AI…
wet bench bath chemical concentration drift prediction, wet bench bath, bench bath chemical, bath chemical concentration, chemical concentration drift
Wet Bench Bath Chemical Concentration Drift Prediction with Machine Learning Introduction & Motivation Wet Bench Bath Chemical Concentration Drift Prediction with Machine Learning…