← Back to AI Factory Chat

AI Factory Glossary

102 technical terms and definitions

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z All
Showing page 2 of 3 (102 entries)

inhomogeneous poisson, time series models

Inhomogeneous Poisson processes have time-varying intensity rates modeling non-stationary event occurrence patterns.

inpainting as pretext, self-supervised learning

Fill masked regions as pretext task.

inpainting diffusion, multimodal ai

Inpainting fills masked regions coherently using diffusion models conditioned on surrounding context.

inpainting mask, generative models

Specify region to modify.

inpainting, generative models

Fill masked regions.

inpainting,generative models

Fill in missing or masked regions of images.

inpainting,outpainting,edit

Inpainting fills masked regions. Outpainting extends beyond borders. AI image editing.

input × gradient, explainable ai

Simple attribution method.

input filter, ai safety

Input filters screen prompts for policy violations or attacks.

input sanitization,ai safety

Clean and validate user inputs before sending to LLM to prevent attacks.

input-dependent depth, model optimization

Input-dependent depth networks use different layer counts per sample based on complexity.

instancenorm, neural architecture

Normalize each sample independently.

instant-ngp, multimodal ai

Instant Neural Graphics Primitives use hash-based encoding for fast NeRF training and rendering.

instruct-pix2pix, multimodal ai

Instruct-pix2pix edits images following natural language instructions.

instructblip,multimodal ai

BLIP fine-tuned for instructions.

instructgpt,foundation model

GPT-3 fine-tuned with RLHF for following instructions.

instruction dataset, training techniques

Instruction datasets contain task descriptions and input-output pairs for supervised learning.

instruction model, llm architecture

Instruction models are fine-tuned to follow natural language commands.

instruction tuning, training techniques

Instruction tuning fine-tunes models on diverse instruction-following tasks.

instructpix2pix,generative models

Image editing following text instructions.

integrated gradients, explainable ai

Path integral of gradients for attribution.

integrated hessians, explainable ai

Use second-order information for attribution.

inter-pair skew, signal & power integrity

Inter-pair skew is delay variation across multiple differential pairs affecting parallel bus timing.

interaction blocks, graph neural networks

Interaction blocks in molecular GNNs update node and edge features through learned message passing.

intercode, ai agents

InterCode evaluates interactive coding agents on execution-based tasks.

interleaved image-text generation,multimodal ai

Generate sequences mixing images and text.

intermediate fusion, multimodal ai

Combine at intermediate layers.

internal failure costs, quality

Costs from defects found before shipping.

internlm,shanghai ai,research

InternLM is Shanghai AI Lab model. Multilingual research model.

interpretability, ai safety

Interpretability makes model decisions understandable to humans.

interpretability,ai safety

Understanding why a model makes specific predictions or decisions.

interpretability,explainability,understand

Interpretability helps understand model decisions. Attention visualization, probing, mechanistic. Build trust.

interpretability,explainability,xai

XAI techniques explain model decisions: attention viz, feature importance, counterfactuals. Important for trust.

interval bound propagation, ibp, ai safety

Compute output bounds for input regions.

intra-pair skew, signal & power integrity

Intra-pair skew is delay mismatch within differential pair reducing common-mode noise immunity.

invariance testing, explainable ai

Test what transformations preserve predictions.

inventory accuracy, supply chain & logistics

Inventory accuracy measures agreement between physical counts and system records.

inverted residual, model optimization

Inverted residuals expand then compress feature maps using depthwise convolutions in between for efficiency.

investor,pitch,deck,fundraising

I can help structure a pitch deck story: problem, solution, traction, market, moat, and financial/roadmap slides.

ion exchange, environmental & sustainability

Ion exchange removes specific ions from water replacing them with hydrogen or hydroxide ions.

ip-adapter, generative models

Image prompt adapter.

ip-adapter, multimodal ai

IP-Adapter enables image prompting by injecting image features into cross-attention layers.

iron-boron pair detection, metrology

Detect Fe-B complexes affecting lifetime.

isolation forest temporal, time series models

Isolation forest adapted for temporal data detects anomalies by measuring path lengths in random decision trees built on windowed features.

isolation forest ts, time series models

Isolation Forest for time series identifies anomalies through path length in random trees over sliding windows.

isotonic regression,ai safety

Non-parametric calibration method.

issue triaging, code ai

Categorize and route bug reports.

iterated amplification, ai safety

Recursively amplify human supervision.

iterated amplification, ai safety

Iterated amplification decomposes oversight into manageable subtasks recursively.

iteration / step,model training

One update of model weights (one batch processed).