Concepts

Concept note는 research note, paper summary, project page, infra note에서 반복해서 쓰는 재사용 가능한 용어를 정의합니다.

이 페이지는 전체 concept map입니다. 처음에는 hub에서 큰 축을 보고, 논문·프로젝트·포스트에서 정확한 정의가 필요할 때 개별 노트로 이동합니다.

주요 허브

• Coverage matrix
• Wiki note quality gate
• Math foundations
• Machine learning
• Data
• Modalities
• Tasks
• Architectures
• Learning methods
• Generative models
• AI systems
• Research methodology
• LLM concepts
• Evaluation
• Geometric deep learning
• Molecular modeling concepts
• Structure-based drug discovery
• Protein modeling concepts
• Genome modeling concepts
• Entities

읽기 경로

• AI 기본기: Math foundations -> Empirical risk minimization -> Generalization -> Stochastic gradient -> Data -> Architectures -> Learning methods -> Augmentation policy -> Evaluation protocol -> Evaluation
• 수학 경로: Linear algebra -> Tensor shape notation -> Calculus -> Matrix calculus -> Probability distribution -> Normal distribution -> Expectation -> Central limit theorem -> Monte Carlo estimation -> Maximum likelihood -> Bayesian inference -> Entropy and KL divergence
• 데이터 경로: Dataset -> Dataset construction checklist -> Example unit -> Split unit -> Dataset split contract -> Preprocessing contract -> Label semantics -> Dataset shift -> Benchmark
• Modality 경로: Modality representation -> Modality-task map -> Text -> Sequence -> Image -> Video -> Audio -> Tabular -> Graph -> 3D structure -> Modality alignment -> Missing modality -> Multimodal learning
• Task 경로: Task specification -> Modality-task map -> Task output space -> Retrieval -> Similarity search -> Reranking -> Question answering -> Sequence generation -> Structured prediction
• Architecture 경로: Inductive bias -> Parameter sharing -> Computational complexity -> Tokenization -> Embedding -> Attention -> Transformer
• Systems 경로: Training run -> Distributed training -> Resource scheduling -> Checkpoint state -> Scaling claim contract -> Inference -> Batch and online inference -> Model serving -> Failure recovery -> Reproducibility
• Research method 경로: Research question -> Hypothesis -> Minimum viable experiment -> Threat to validity -> Result interpretation
• Agent 경로: Agent architecture -> Agent environment -> Action space -> Tool contract -> Tool result handling -> Acceptance criteria -> Verification loop
• Graph와 geometry: Geometry -> Symmetry group -> Coordinate frame -> Distance geometry -> Graph construction -> Graph neural networks -> Invariant feature -> Equivariant feature -> Equivariant GNN
• Structure-based modeling: Pocket -> Binding site -> Pocket representation -> Receptor and ligand preparation -> Protein-ligand interaction -> Docking workflow -> Pose generation -> Pose quality -> Scoring function -> Binding affinity -> Virtual screening
• Molecular modeling: Molecular standardization -> Molecular identity -> Chemical state contract -> Molecular featurization contract -> SMILES -> Molecular graph -> Molecular fingerprint -> Molecular similarity -> Molecular property prediction -> Conformer -> Force field
• Protein modeling: Sequence -> Protein domain -> Protein representation -> Contact map -> Protein structure prediction
• Genome-level sequence: Genome -> Genomic region -> K-mer -> Variant effect prediction
• 논문 읽기: Paper triage -> Paper note format -> Claim extraction -> Result table reading -> Benchmark card -> Ablation map -> Reproduction plan -> Reading status
• Coverage 확인: Wiki note quality gate -> Coverage matrix -> AI paper intake -> Computational Biology paper intake -> Formula intake -> Benchmark intake -> AI Computational Biology Math post intake
• LLM Wiki: Language model -> Context window -> Token budget -> Context packing -> Chunking -> Query rewriting -> Embedding retrieval -> Hybrid retrieval -> Retrieval-augmented generation -> Evidence-grounded generation -> Prompt injection boundary

Entity Concepts

• Entities
• Entity relation map
• Target
• Protein
• Pocket
• Ligand
• Molecule
• Protein-ligand complex
• Assay
• Bioactivity label
• Dataset

Math Foundations

• Math foundations
• Linear algebra
• Tensor shape notation
• Eigenvalue and eigenvector
• Singular value decomposition
• Calculus
• Matrix calculus
• Geometry
• Symmetry group
• Random variable
• Probability distribution
• Normal distribution
• Expectation
• Covariance and correlation
• Central limit theorem
• Monte Carlo estimation
• Maximum likelihood
• Entropy and KL divergence
• Bayes rule
• Bayesian inference
• Statistical estimator
• Hypothesis testing
• Bias-variance tradeoff

Data

• Data
• Coverage matrix
• Dataset
• Dataset construction checklist
• Dataset card
• Example unit
• Split unit
• Dataset split contract
• Data distribution
• Data schema
• Preprocessing contract
• Data curation
• Data versioning
• Annotation and labeling
• Label semantics
• Label noise
• Missing data
• Censored label
• Weak label
• Class imbalance
• Dataset shift
• Sampling strategy
• Sampling bias
• Metadata and provenance
• Benchmark
• Evaluation set design
• Data preprocessing

Modalities

• Modalities
• Modality representation
• Modality-task map
• Text
• Sequence
• Image
• Video
• Audio
• Tabular
• Graph
• 3D structure
• Modality alignment
• Missing modality
• Multimodal learning

Tasks

• Tasks
• Task specification
• Task output space
• Modality-task map
• Classification
• Regression
• Ranking
• Property prediction
• Interaction prediction
• Retrieval
• Similarity search
• Reranking
• Object detection
• Localization
• Segmentation
• Captioning
• Question answering
• Sequence generation
• Structured prediction
• Coordinate prediction
• Graph prediction
• Time-series forecasting
• Anomaly detection

Architectures

• Architectures
• Inductive bias
• Parameter sharing
• Architecture selection
• Computational complexity
• Linear layer
• Weight initialization
• Activation function
• Feed-forward network
• Normalization
• Normalization placement
• Residual connection
• Tokenization
• Embedding
• Positional encoding
• Attention
• Cross-attention
• Pooling and readout
• MLP
• Convolution
• CNN
• Residual network
• U-Net
• RNN
• Graph construction
• Graph neural networks
• Deep Sets
• Transformer
• Encoder-only Transformer
• Decoder-only Transformer
• Vision Transformer
• State-space models
• Mamba
• Mixture of experts

Machine Learning

• Machine learning
• Data preprocessing
• Feature engineering
• Probabilistic prediction
• Decision rule
• Classification
• Regression
• Ranking
• Density estimation
• Representation learning
• Representation evaluation
• Linear probing
• Empirical risk minimization
• Generalization
• Overfitting and underfitting
• Model selection
• Hyperparameter tuning
• Early stopping
• Learning curve
• Validation curve
• Loss function
• Cross-entropy loss
• Mean squared error
• Negative log-likelihood
• Training loop
• Stochastic gradient
• Gradient descent
• Backpropagation
• Optimizer
• Second-order optimization
• Adam
• AdamW
• Learning rate schedule
• Training stability
• Training step accounting
• Model state contract
• Weight decay
• Gradient clipping
• Gradient accumulation
• Batch size
• Linear model
• Tree-based model
• Kernel method
• Ensemble method
• Regularization
• Optimization
• Dimensionality reduction
• test split
• Leakage
• Dataset shift

Learning Methods

• Learning methods
• Supervised learning
• Semi-supervised learning
• Pretraining
• Self-supervised learning
• Representation evaluation
• Linear probing
• Masked modeling
• Contrastive learning
• Augmentation policy
• Representation collapse
• JEPA
• Transfer learning
• Fine-tuning
• Fine-tuning protocol
• Knowledge distillation
• Instruction tuning
• Domain adaptation
• Curriculum learning
• Reinforcement learning
• Preference optimization

Structure-Based Modeling

• Structure-based drug discovery
• Receptor and ligand preparation
• Pose generation
• Protein-ligand interaction
• Docking workflow
• Scoring function
• Pose quality
• Interaction fingerprint
• Binding affinity
• Virtual screening
• Protein-ligand split
• Template leakage

Computational Biology Concepts

• Molecular modeling concepts
• Molecular standardization
• Molecular featurization contract
• SMILES
• Molecular graph
• Molecular fingerprint
• Molecular similarity
• Molecular property prediction
• Substructure search
• Fragment-SELFIES
• Conformer
• Force field
• Energy minimization
• Molecular dynamics
• Tautomer
• Protonation state
• Stereochemistry

Protein Modeling

• Protein modeling concepts
• Protein representation
• Protein domain
• Binding site
• Pocket representation
• Protein structure prediction
• Multiple sequence alignment
• Sequence identity clustering
• Contact map
• Sequence-structure alignment

Genome Modeling

• Genome modeling concepts
• Genomic region
• K-mer
• Variant effect prediction
• Genome annotation

Generative Models

• Generative models
• Latent variable model
• Evidence lower bound
• Conditional generation
• Sampling
• Guidance
• Autoregressive model
• Diffusion model
• Score-based model
• Score matching
• Probability flow ODE
• Flow matching
• Rectified flow
• Normalizing flow
• VAE
• GAN
• Molecular generation
• Protein design

AI Systems

• AI systems
• Training run
• Distributed training
• Resource scheduling
• Environment management
• Inference
• Inference contract
• Model card
• Model versioning
• Data validation
• Model serving
• Deployment strategy
• Latency and throughput
• Memory-compute tradeoff
• Storage and IO
• Observability
• Experiment tracking
• Reproducibility

Research Methodology

• Research methodology
• Research question
• Hypothesis
• Experiment design
• Minimum viable experiment
• Result interpretation
• Threat to validity
• Research log

LLM Concepts

• LLM concepts
• Language model
• Context window
• Token budget
• Context packing
• Tool calling
• Prompt injection boundary
• Prompting
• In-context learning
• Decoding
• Structured output
• Hallucination and grounding
• Embedding retrieval
• Chunking
• Hybrid retrieval
• Query rewriting
• Retrieval-augmented generation
• Agent architecture
• Agent orchestration

Geometric Deep Learning

• Geometric deep learning
• Geometric architecture
• Equivariant GNN

Learning Methods

• Learning methods
• Self-supervised learning
• Masked modeling
• Representation evaluation
• Linear probing
• JEPA
• Contrastive learning
• Fine-tuning
• Knowledge distillation
• Transfer learning
• Fine-tuning protocol
• Preference optimization

Evaluation

• Evaluation
• Evaluation protocol
• Evaluation set design
• Metric
• Metric selection
• Failure mode taxonomy
• Confusion matrix
• Threshold selection
• Classification metrics
• Probability metrics
• Regression metrics
• Ranking metrics
• Generation evaluation
• Baseline
• Ablation study
• Confidence interval
• Seed variance
• Bootstrap evaluation
• Statistical significance
• Effect size
• Paired comparison
• Multiple comparisons
• test split
• Cross-validation
• Leakage
• Test-set contamination
• Benchmark saturation
• Applicability domain
• Assay harmonization
• Negative set
• Activity cliff
• Scaffold split
• Protein family split
• Protein-ligand split
• OOD generalization
• Calibration
• Reliability diagram
• Uncertainty estimation
• Robustness
• Error analysis
• Interpretability

Research Links

• Structure-based modeling
• Protein modeling
• Generative models