27,049 technical terms and definitions
speculative decoding kv cache optimization, speculative decoding, kv cache optimization, gb gpu gqa kv llm mha, speculative decoding & kv-cache optimization, speculative decoding kv-cache
Speculative Decoding & KV-Cache Optimization Introduction & Motivation Autoregressive language model inference is fundamentally sequential: generating each new token requires…
draft model verification, parallel token generation, speculative sampling inference, assisted generation
Speculative Decoding is the LLM inference acceleration technique that uses a small, fast "draft" model to generate multiple candidate tokens in parallel, which the large "target"…
draft model verification, speculative sampling, llm inference acceleration, assisted generation
Speculative Decoding is the LLM inference acceleration technique that uses a smaller, faster "draft" model to generate candidate token sequences speculatively, then verifies them…
draft model verification, parallel token generation, speculative sampling inference, assisted generation
Speculative Decoding is the inference acceleration technique that uses a small draft model to generate multiple candidate tokens in parallel, then verifies them with the target…
speculative task execution, mapreduce speculative launch, distributed recovery acceleration, tail tolerance compute
Speculative Execution in Distributed Systems is the execution strategy that runs backup copies of uncertain tasks to reduce completion time variance. What It Covers Core concept…
thread speculation, speculative parallelism, optimistic execution, spec thread
Speculative Execution in Parallel Systems is the technique of optimistically executing tasks in parallel before knowing whether their results will be needed — gambling that…
thread level speculation, optimistic concurrency, speculative parallelism, hardware speculation
Speculative Execution and Parallelism is the hardware and software technique that optimistically executes computation before its preconditions are confirmed — overlapping…
speculative execution vulnerability mitigation, speculative execution vulnerability, execution vulnerability mitigation, speculative execution, execution vulnerability
Speculative Execution Vulnerability Mitigation Introduction & Motivation Speculative Execution Vulnerability Mitigation addresses a central problem in security process control…
code smell, yagni
Speculative Generality is a code smell where abstractions, interfaces, or framework-like structures are created for anticipated future needs that never materialize. What Is…
txn memory, speculative execution, thread speculation
Speculative Parallelism and Transactional Memory are techniques for extracting parallelism from code with potential data dependencies by optimistically executing tasks in parallel…
optimization
Speculative Sampling is a decoding strategy where a draft model proposes tokens and a stronger model verifies them - It is a core method in modern semiconductor AI serving…
speculative decoding, draft model, target verification, acceptance sampling, self speculation, medusa, lookahead decoding
Speculative sampling is an exact acceleration method in which a faster draft process proposes several future tokens and a target model verifies them in parallel. When draft tokens…
nlp
Speech act recognition is classification of utterance function such as request question promise or statement - Recognition models map language patterns and context…
audio & speech
Speech-Driven Gestures is generation of body and hand gestures conditioned on spoken audio or text. - It models co-speech motion so avatars and agents express natural nonverbal…
speech emotion recognition, speech emotion, emotion recognition
Speech Emotion Recognition Introduction & Motivation Speech Emotion Recognition addresses a central problem in nlp process control and emotion-detection-voice-acoustic prediction…
Speech Emotion Recognition Introduction & Motivation Speech Emotion Recognition addresses a central problem in multimodal learning and emotion-recognition-accuracy alignment: how…
speech enhancement denoising, speech enhancement, enhancement denoising
Speech Enhancement Denoising Introduction & Motivation Speech Enhancement Denoising addresses a central problem in nlp process control and speech-enhancement-noise-reduction…
audio language model, audiopalm, whisper, speech ai foundation
Speech Language Models are the foundation models that process and generate speech directly as a native modality — either by tokenizing audio into discrete units that language…
keyword spotting chip, neural engine voice, always on audio processor, wake word detection chip
Speech and Audio Processing Chip: Always-On Keyword Spotting Engine — ultra-low-power neural network for wake-word detection enabling voice assistant activation with <1 mW standby…
automatic speech recognition, asr, whisper, ctc speech, wav2vec
speech recognition is automatic speech recognition that converts acoustic waveforms into text with timestamps, language, and confidence. ASR powers assistants, captions, call…
speech recognition, recognition acoustic, ctc dct mfcc passed rnn timit, speech recognition & acoustic modeling, speech recognition acoustic, recognition acoustic modeling, acoustic modeling
Speech Recognition & Acoustic Modeling Introduction & Motivation Speech Recognition: convert audio to text. Acoustic models, language models, end-to-end systems. Applications…
whisper speech model, connectionist temporal classification ctc, end to end speech, automatic speech recognition
Automatic Speech Recognition (ASR) is the deep learning system that converts spoken audio into text — processing raw audio waveforms through neural encoder-decoder architectures…
whisper speech model, conformer asr architecture, ctc attention hybrid, end to end speech recognition
Speech Recognition (ASR) Transformers are neural architectures that convert spoken audio into text by processing mel-spectrogram features through encoder-decoder or encoder-only…
speech recognition, audio representation learning, asr cer ctc gpu lium llm, speech recognition & audio representation learning, speech recognition audio, recognition audio representation
Speech Recognition & Audio Representation Learning Introduction & Motivation Automatic speech recognition (ASR) and audio representation learning form the technological foundation…
speech recognition voice control, speech recognition voice, recognition voice control, speech recognition, recognition voice, voice control
Speech Recognition Voice Control Introduction & Motivation Speech Recognition Voice Control addresses a central problem in robotics process control…
text to speech neural, wavenet vocoder, tacotron mel spectrogram, neural speech generation
Neural Text-to-Speech (TTS) is the deep learning pipeline that converts text into natural-sounding speech waveforms — typically through a two-stage architecture where an acoustic…
Text-to-Speech Synthesis Introduction & Motivation Text-to-Speech Synthesis addresses a central problem in generative generation and speech-naturalness-intelligibility synthesis…
speech text multimodal fusion, speech text multimodal, text multimodal fusion, speech text, text multimodal, multimodal fusion
Speech Text Multimodal Fusion Introduction & Motivation Speech Text Multimodal Fusion addresses a central problem in nlp process control…
speech to text asr, speech to text, to text asr, speech to, to text, text asr
Speech to Text ASR Introduction & Motivation Speech to Text ASR addresses a central problem in nlp process control and asr-word-error-rate-metric prediction: how to Predict…
speech-to-text, stt / asr, audio
Speech-to-text (STT), also known as Automatic Speech Recognition (ASR), transcribes spoken audio into written text, converting acoustic signals into sequences of words. ASR is…
speed power binning optimization machine learning, speed power, binning optimization machine, speed and power binning optimization, speed power binning, power binning optimization
Speed and Power Binning Optimization with Machine Learning Introduction & Motivation Speed and Power Binning Optimization with Machine Learning addresses a central problem…
business
Speed binning is the practice of testing each die and sorting by maximum operating frequency, then selling faster chips at premium prices and slower chips at lower prices. It's…
serdes phy design, high-speed interface, pcie phy, ddr phy
High-Speed Interface PHY Design (SerDes, PCIe, DDR) is the mixed-signal circuit design discipline focused on creating the physical-layer transceivers that reliably transmit…
I/O high-speed, equalization techniques, signal integrity
High-Speed I/O Equalization and Signal Integrity Techniques is methods correcting channel-induced signal degradation enabling reliable data transfer over limited-bandwidth…
manufacturing operations
Speed Loss is output reduction caused by operating below ideal cycle speed during runtime - It erodes performance even when equipment is technically running. What Is Speed Loss?…
audio & speech
Speed Perturbation is speech augmentation by resampling audio to simulate faster or slower speaking rates - It increases speaker and prosody diversity without collecting new…
audio & speech
SpeedySpeech is a non-autoregressive TTS architecture for low-latency mel-spectrogram generation. - It predicts speech frames in parallel to reduce inference time significantly.…
supply chain & logistics
Spend Analysis is systematic analysis of procurement spending patterns across suppliers, categories, and regions - It reveals savings opportunities, compliance gaps…
sphereface angular softmax and angular constraint embedding, sphereface angular softmax, angular softmax angular, softmax angular constraint, angular constraint embedding
SphereFace Angular Softmax and Angular Constraint Embedding Introduction & Motivation SphereFace Angular Softmax and Angular Constraint Embedding addresses a central problem…
graph neural networks
SphereNet is a three-dimensional molecular graph network modeling distances angles and torsions. - It captures full local geometry including chirality-sensitive spatial…
computer vision
Spherical CNNs are neural networks that perform convolution directly on the surface of a sphere ( ) rather than on flat image planes — using either spectral methods (spherical…
graph neural networks
Spherical Harmonics is orthogonal basis functions on the sphere used to encode angular dependence in 3D graph models - They provide a mathematically grounded angular decomposition…
sh, 3d vision
Spherical harmonics for color is the basis-function representation that models view-dependent color variation as coefficients over angular functions - it provides an efficient way…
serial peripheral interface, sclk, mosi, miso, chip select, qspi, ospi, spi bus
SPI protocol is a synchronous serial interface in which a controller supplies clock and chip select while exchanging data over separate output and input lines. Its simple…
spi slave hold time margin prediction, spi slave hold, slave hold time, hold time margin, time margin prediction, spi slave, slave hold, hold time, time margin, margin prediction
SPI Slave Hold Time Margin Prediction with Machine Learning Introduction & Motivation SPI Slave Hold Time Margin Prediction with Machine Learning addresses a central problem…
spice, signal & power integrity
SPICE simulation is circuit-level simulation using device and interconnect models to predict analog and mixed-signal behavior - Numerical solvers evaluate transient, AC, and DC…
spice simulation accuracy prediction, spice simulation accuracy, simulation accuracy prediction, spice simulation, simulation accuracy, accuracy prediction
SPICE Simulation Accuracy Prediction with Machine Learning Introduction & Motivation SPICE Simulation Accuracy Prediction with Machine Learning addresses a central problem…
spice, simulation program with integrated circuit emphasis, design
SPICE (Simulation Program with Integrated Circuit Emphasis) Overview SPICE is the standard circuit simulation engine used to verify analog, digital, and mixed-signal IC designs…
Spike-and-Slab Priors for Exact Bayesian Feature Selection Introduction & Motivation Spike-and-Slab Priors for Exact Bayesian Feature Selection addresses a central problem…
implant, dopant activation rapid thermal millisecond, junction depth minimization crystal damage repair, RTA spike temperature profile dopant
Spike annealing activates implanted dopants and repairs ion-implantation crystal damage while minimizing the diffusion that would otherwise smear a shallow junction into a deeper…