EXPLORE / 06

Terms index

Specialized AI and software-development terms with concise explanations

Entity records

4490 records · 35/113

  1. 1361

    Permission Isolation

    A security mechanism restricting users or processes to access only authorized data and resources within a software system.

    Website pending
  2. 1362

    Adversarial Prompt Injection

    A security attack where crafted inputs manipulate model decisions or induce unintended actions.

    Website pending
  3. 1363

    Adversarial Machine Learning

    A field studying how attackers manipulate inputs or model behavior to compromise systems, and how defenders build robust models

    Website pending
  4. 1364

    Pitch Correction

    An audio processing technique adjusting vocal or instrumental pitch to match a key, often categorized as post-production rather

    Website pending
  5. 1365

    Defect Petition

    A legal procedure where the public requests regulators to investigate specific product safety defects, often preceding formal re

    Website pending
  6. 1366

    Robotaxi

    Commercial passenger vehicles operating at L4/L5 autonomy levels without human driver intervention.

    Website pending
  7. 1367

    Decoding Compute

    The computational cost incurred during autoregressive output generation, primarily driven by sequence length (token count), serv

    Website pending
  8. 1368

    Latent Space Reasoning

    A technique performing multi-step logical derivation within the model's continuous vector representations, avoiding intermediate

    Website pending
  9. 1369

    Data Brokerage

    Business practice of collecting, cleaning, and packaging proprietary or niche data to sell as training corpora or RAG sources to

    Website pending
  10. 1370

    τ-Voice Benchmark

    A benchmark evaluating voice assistants' ability to execute complex tasks in real-world scenarios, involving tool calling, polic

    Website pending
  11. 1371

    FP4 Compute

    Matrix operations using 4-bit floating-point precision to significantly boost inference throughput and reduce power consumption,

    Website pending
  12. 1372

    HBM3E

    Third-generation enhanced High Bandwidth Memory providing extreme data transfer rates to alleviate memory bandwidth bottlenecks

    Website pending
  13. 1373

    Agentic Inference Benchmark

    A benchmark that replays real agentic coding sessions, including prompt caching and tool-call delays, to measure realistic infer

    Website pending
  14. 1374

    Speech-to-Speech Model

    End-to-end architecture that directly processes audio input to generate audio output, bypassing text intermediaries to reduce la

    Website pending
  15. 1375

    Catalytic Core

    Minimal essential structural domain in an enzyme directly involved in substrate binding and bond breaking/forming, determining s

    Website pending
  16. 1376

    De Novo Protein Design

    Computational method generating amino acid sequences from scratch based on functional/structural constraints without natural tem

    Website pending
  17. 1377

    Read-Copy-Update (RCU)

    A synchronization mechanism in the Linux kernel allowing lock-free reads, where writers create copies and delay freeing old data

    Website pending
  18. 1378

    Content Curation

    The process of selecting, organizing, and presenting high-quality relevant content from vast sources to specific audiences.

    Website pending
  19. 1379

    SRAM

    Volatile memory retaining data without refresh, used in caches; its scaling directly impacts chip data throughput efficiency and

    Website pending
  20. 1380

    Nanosheet Transistor

    A gate-all-around FET structure stacking horizontal silicon nanosheets vertically to increase transistor density and current dri

    Website pending
  21. 1381

    Validation Bottleneck

    The phenomenon where AI content generation speed outpaces human expert review capacity, creating a trust and efficiency barrier

    Website pending
  22. 1382

    Intelligent Document Processing (IDP)

    A technology stack combining OCR, NLP, and ML to automatically extract key information from unstructured or semi-structured docu

    Website pending
  23. 1383

    Cross-modal Reasoning

    The ability to integrate and logically derive insights from heterogeneous data sources, such as images, text, and sensor reading

    Website pending
  24. 1384

    Geospatial Foundation Models

    Large-scale models pre-trained on earth science data (remote sensing, weather, hydrology) to extract spatiotemporal features for

    Website pending
  25. 1385

    Disaggregated Inference

    An architectural pattern deploying compute-intensive prefill and memory-bandwidth-sensitive decode phases on separate hardware i

    Website pending
  26. 1386

    Multimodal Scaling Laws

    Empirical rules quantifying performance gains with increased parameters, data, and compute, specifically focusing on how modalit

    Website pending
  27. 1387

    Stochastic Differential Equation

    A mathematical framework describing dynamic systems with random terms, used in deep learning to model noise schedules or stochas

    Website pending
  28. 1388

    Masked Diffusion Model

    A variant of discrete diffusion models that generates data by randomly masking tokens and iteratively predicting masked parts, o

    Website pending
  29. 1389

    Frame-Level Activation

    Neuron output values generated by speech models for each time frame, reflecting the encoded state of local acoustic features at

    Website pending
  30. 1390

    Hidden State Dimension

    Specific components of the vector representation output by intermediate neural network layers, used here to capture acoustic sig

    Website pending
  31. 1391

    Self-Replication Mechanism

    Behavioral pattern where software automatically creates copies of itself and installs them on other systems, common in malware a

    Website pending
  32. 1392

    Cloud Inference

    Deploying compute-intensive LLM inference on servers while devices handle I/O, suitable for resource-constrained endpoints requi

    Website pending
  33. 1393

    Hybrid Modeling

    An architectural paradigm combining physical mechanism models' data-driven advantages with ML's statistical fitting capabilities

    Website pending
  34. 1394

    Spatial Downscaling

    Mapping low-resolution large-scale meteorological data to high-resolution local details using algorithms to capture small-scale

    Website pending
  35. 1395

    Climate Emulator

    A machine learning-based surrogate model approximating physical climate simulations to reduce computational costs and accelerate

    Website pending
  36. 1396

    Chain-of-Thought Monitoring

    An interpretability technique analyzing intermediate reasoning steps to detect unsafe behaviors or logical errors, crucial for A

    Website pending
  37. 1397

    Personalized Recommendation System

    A mechanism that predicts user interests and pushes customized content or ads using algorithms based on historical behavior, pre

    Website pending
  38. 1398

    Causal Attention

    The core component of Transformer decoders that restricts each token to attend only to itself and preceding tokens, ensuring log

    Website pending
  39. 1399

    Prefix Cache

    A system-level optimization allowing different requests to share computation results for identical prefixes (e.g., System Prompt

    Website pending
  40. 1400

    Retrieval-Augmented Generation (RAG)

    A technique combining external knowledge base retrieval with LLM generation to reduce hallucinations and improve accuracy by inj

    Website pending