🔬 Applied Scientist & ML Specialist938 Problems · 127 Topics🌐 Linked to Tech Vault

TalentMe Science Depth: Universal Applied Science & ML Depth Problem Bank

938 end-to-end science depth, system design, transition bridge & BQ problems covering derivations, production trade-offs, pitfalls, and follow-ups.

🎯

Science Depth Mastery Tracker

0% Mastered

Locally synced in your browser. Track, review, and master all 938 scientific & design topics.

Mastered: 0
Reviewing: 0
Unstarted: 89
🚀 Start Next Unpracticed →
🌐
Linked to TalentMe Tech Vault & MindmapsM1: Mathematics & Statistics Fundamentals
All 938 questions are directly linked to TalentMe's Tech Vault guides, atomic cards, and interactive mindmaps.
🧭Math Foundations Mindmap & Cards→📖Bayesian Inference, Entropy & KL Divergence→🎯DS Track→
Found 24 in-depth questions
M1-025Mathematics & Statistics FundamentalsCalculus & Taylor Expansion
Hard

Explain Second-Order Sufficient Conditions and Saddle Points, and Why Deep Learning Escapes Saddle Points

⚡ One-Sentence Answer

A positive definite Hessian guarantees a strict local minimum, while an indefinite Hessian signifies a saddle point; in high-dimensional loss landscapes, saddle points vastly outnumber local minima, but SGD gradient noise escapes them efficiently.

Math & Tradeoffs→
M1-026Mathematics & Statistics FundamentalsConvex Optimization & KKT
Easy

Define Convex Sets and Convex Functions, and Explain Why Convexity Guarantees Global Optimality

⚡ One-Sentence Answer

A set is convex if all line segments connecting any two points lie inside the set; a function is convex if chords lie above the graph, guaranteeing that any local minimum is a global minimum.

Math & Tradeoffs→
M1-027Mathematics & Statistics FundamentalsConvex Optimization & KKT
Medium

Formulate the Karush-Kuhn-Tucker (KKT) Conditions and Explain the Physical Meaning of Complementary Slackness

⚡ One-Sentence Answer

KKT conditions are necessary first-order optimality criteria for constrained optimization, with complementary slackness $\lambda_i g_i(x)=0$ dictating that inactive constraints have zero shadow price.

Math & Tradeoffs→
M1-028Mathematics & Statistics FundamentalsConvex Optimization & KKT
Medium

Define Strong Duality and Slater's Condition, and Explain the Practical Utility of Dual Formulations

⚡ One-Sentence Answer

Weak duality guarantees dual bound $d^* \le p^*$; strong duality ($d^* = p^*$) holds for convex problems satisfying Slater's condition (strictly feasible interior point), enabling solving complex primal problems via simpler duals.

Math & Tradeoffs→
M1-029Mathematics & Statistics FundamentalsConvex Optimization & KKT
Medium

Explain Geometrically and Algebraically Why L1 Regularization Promotes Sparsity While L2 Does Not

⚡ One-Sentence Answer

Geometrically, L1 contour balls have sharp non-smooth corners on coordinate axes where loss contours touch first; algebraically, L1 subgradient produces a constant shrinkage threshold, setting small weights exactly to zero.

Math & Tradeoffs→
M1-030Mathematics & Statistics FundamentalsConvex Optimization & KKT
Hard

Explain Subgradients and Proximal Operators, and How Proximal Gradient Descent Optimizes Non-Smooth L1 Penalties

⚡ One-Sentence Answer

A subgradient generalizes derivatives for non-smooth convex functions via supporting hyperplanes, while proximal operators solve composite optimization via exact closed-form soft-thresholding.

Math & Tradeoffs→
M1-031Mathematics & Statistics FundamentalsNumerical Stability
Easy

Explain Why Softmax Must Subtract the Maximum Value and How It Prevents Numerical Overflow

⚡ One-Sentence Answer

Subtracting $x_{\max}$ prevents $e^{x_i}$ from exceeding floating-point representation limits ($+\infty$), exploiting shift invariance: $\text{Softmax}(x) = \text{Softmax}(x - c)$.

Math & Tradeoffs→
M1-032Mathematics & Statistics FundamentalsNumerical Stability
Easy

Formulate the Log-Sum-Exp (LSE) Trick and Explain What Numerical Instability It Resolves

⚡ One-Sentence Answer

The Log-Sum-Exp trick computes $\log\sum_i e^{x_i} = c + \log\sum_i e^{x_i - c}$ using $c = \max_i x_i$, preventing catastrophic overflow from large positive values and underflow from large negative values.

Math & Tradeoffs→
M1-033Mathematics & Statistics FundamentalsNumerical Stability
Medium

Explain Loss Scaling in Mixed Precision (FP16/BF16) Training and How It Prevents Underflow

⚡ One-Sentence Answer

Loss scaling multiplies the forward loss by a scale factor $S$ before backprop, shifting small gradient values into the representable dynamic range of FP16 to prevent them from flushing to zero.

Math & Tradeoffs→
M1-034Mathematics & Statistics FundamentalsNumerical Stability
Medium

Explain Why Numerically Stable Implementations Use log1p(exp(-|z|)) for Binary Cross-Entropy (BCEWithLogits)

⚡ One-Sentence Answer

The formulation $\text{BCE}(z, y) = \max(z, 0) - yz + \log(1 + e^{-|z|})$ guarantees numerical stability for both large positive and negative logits $z$, avoiding overflow in $\exp(z)$ and loss of precision in $\log(1+x)$.

Math & Tradeoffs→
M1-035Mathematics & Statistics FundamentalsNumerical Stability
Hard

Enumerate Common Sources of Numerical Instability in Deep Learning and Their Standard Industrial Mitigations

⚡ One-Sentence Answer

Numerical instability stems from vanishing/exploding gradients, floating-point overflow/underflow, matrix near-singularity, and precision loss; mitigated by normalization, gradient clipping, log-domain computation, and mixed precision design.

Math & Tradeoffs→
M1-036Mathematics & Statistics Fundamentals估计理论 (MLE/MAP)
Easy

Define Maximum Likelihood Estimation (MLE) with a Complete End-to-End Derivation

⚡ One-Sentence Answer

MLE finds parameter values that maximize the probability of generating the observed sample dataset: $\hat{\theta}_{\text{MLE}} = \arg\max_\theta \sum_{i=1}^N \log p(x_i\mid \theta)$.

Math & Tradeoffs→
M1-037Mathematics & Statistics Fundamentals估计理论 (MLE/MAP)
Easy

Define Maximum A Posteriori (MAP) Estimation and Explain Its Exact Connection to MLE and Regularization

⚡ One-Sentence Answer

MAP maximizes the posterior mode: $\hat{\theta}_{\text{MAP}} = \arg\max [\log p(D\mid \theta) + \log p(\theta)]$; a Gaussian prior corresponds to L2 regularization (Ridge), while a Laplace prior corresponds to L1 (Lasso).

Math & Tradeoffs→
M1-038Mathematics & Statistics Fundamentals估计理论 (MLE/MAP)
Medium

Explain the Bias-Variance Decomposition and Write the Decomposition for Expected Generalization Error

⚡ One-Sentence Answer

Expected prediction error decomposes into Bias squared (underfitting from wrong model assumptions), Variance (overfitting from sensitivity to training data noise), and Irreducible Error (inherent label noise).

Math & Tradeoffs→
M1-039Mathematics & Statistics Fundamentals估计理论 (MLE/MAP)
Medium

Define Statistical Consistency and Asymptotic Efficiency, and Assess Whether MLE Satisfies Both

⚡ One-Sentence Answer

Consistency means estimates converge in probability to the true parameter as $N\to\infty$; efficiency means achieving the minimum possible variance (Cramér-Rao bound); MLE satisfies both under standard regularity conditions.

Math & Tradeoffs→
M1-040Mathematics & Statistics Fundamentals估计理论 (MLE/MAP)
Hard

Define Model Identifiability and Provide a Concrete Example of Non-Identifiability in Latent Variable Models

⚡ One-Sentence Answer

A model is identifiable if distinct parameter values produce distinct probability distributions: $\theta_1 \ne \theta_2 \implies P_{\theta_1} \ne P_{\theta_2}$; without identifiability, parameters cannot be uniquely learned from data.

Math & Tradeoffs→
M1-041Mathematics & Statistics FundamentalsHypothesis Testing
Easy

Explain the Precise Definition of p-value and Clarify Its Most Prevalent Misconceptions

⚡ One-Sentence Answer

The p-value is the probability of observing a test statistic at least as extreme as the empirical result, assuming the null hypothesis $H_0$ is true: $P(T \ge t_{\text{obs}} \mid H_0)$; it is NOT the probability that $H_0$ is true.

Math & Tradeoffs→
M1-042Mathematics & Statistics FundamentalsHypothesis Testing
Easy

Define Type I Error, Type II Error, Significance Level (Alpha), and Statistical Power (Beta)

⚡ One-Sentence Answer

Type I error $\alpha$ is false positive (rejecting true $H_0$); Type II error $\beta$ is false negative (failing to reject false $H_0$); Power $1-\beta$ is the probability of correctly detecting a real effect.

Math & Tradeoffs→
M1-043Mathematics & Statistics FundamentalsHypothesis Testing
Medium

Explain the Mathematical Relationship Between Sample Size and Statistical Power with Sizing Formulas

⚡ One-Sentence Answer

Sample size scales inversely with the square of Minimum Detectable Effect ($N \propto 1/\delta^2$); halving the effect size requires quadrupling the sample size.

Math & Tradeoffs→
M1-044Mathematics & Statistics FundamentalsHypothesis Testing
Medium

Define the Multiple Testing Problem and Contrast Common Corrections (Bonferroni vs. Benjamini-Hochberg FDR)

⚡ One-Sentence Answer

Testing multiple hypotheses simultaneously inflates family-wise false positives exponentially: $1 - (1-\alpha)^m$; controlled via conservative Bonferroni (FWER) or powerful Benjamini-Hochberg (FDR).

Math & Tradeoffs→
M1-045Mathematics & Statistics FundamentalsHypothesis Testing
Hard

Explain Sequential Testing and the Peeking Problem, and Detail Correct Continuous Monitoring Frameworks

⚡ One-Sentence Answer

The Peeking Problem occurs when continuously checking p-values and stopping once $p<0.05$, which inflates false positive rates from 5% to over 30%; resolved via alpha spending or sequential probability ratio tests.

Math & Tradeoffs→
M1-046Mathematics & Statistics FundamentalsConfidence Intervals & Bootstrap
Easy

Explain the Frequentist Interpretation of Confidence Intervals and Their Duality with Hypothesis Testing

⚡ One-Sentence Answer

A 95% confidence interval means that across repeated hypothetical experiments, 95% of constructed intervals contain the fixed true parameter $\theta$; hypothesis testing and confidence intervals are mathematically dual.

Math & Tradeoffs→
M1-047Mathematics & Statistics FundamentalsConfidence Intervals & Bootstrap
Easy

Explain the Non-Parametric Bootstrap Principle and Outline the Exact Procedure for Mean Confidence Intervals

⚡ One-Sentence Answer

The bootstrap approximates the unknown population distribution using the empirical sample distribution, simulating sampling variance by resampling with replacement $B$ times from the observed data.

Math & Tradeoffs→
M1-048Mathematics & Statistics FundamentalsConfidence Intervals & Bootstrap
Medium

Compare the Three Standard Bootstrap Interval Construction Methods (Percentile, Normal, and BCa Intervals)

⚡ One-Sentence Answer

Standard Normal bootstrap assumes asymptotic symmetry, Percentile bootstrap uses empirical quantiles directly, and BCa (Bias-Corrected and Accelerated) corrects for both median bias and variance skewness.

Math & Tradeoffs→