ACCT 331: WEEK 9

ACCT 331: INTRODUCTION TO APPLIED ARTIFICIAL INTELLIGENCE

Course Information

  • Location: Schreiber #302

  • Schedule: Tuesday & Thursday, 1:00-2:15

  • Course Code: 10304

  • Additional Notes: www.company.com

MATHEMATICS IN ARTIFICIAL INTELLIGENCE

  • AI is firmly rooted in mathematics:

    • It is not magic, but rather mathematics.

    • Emphasis on quantitative methods used to model, predict, and interpret data.

NEWS YOU CAN USE

Walmart and OpenAI Collaboration

  • Partnership Overview:

    • Walmart partners with OpenAI to enhance shopper experiences.

  • Key Insights:

    • Customers can make purchases within ChatGPT.

    • AI-first shopping moves from reactive to proactive.

    • Enhancements in product catalogues and customer service.

    • Educational initiatives for associates to promote AI literacy.

JPMorgan Chase AI Research Associate Program - Internship

  • Job Information:

    • Position: AI Research Associate

    • Location: New York, NY (383 Madison Ave, New York, NY, 10179, US)

    • Salary Range: $135,000.00-$155,000.00

  • Description:

    • Focus on cutting-edge research involving AI, machine learning (ML), and cryptography.

    • Collaborate with teams to integrate AI solutions in finance.

Data Science Internship at IBM Quantum

  • Opportunity: Summer internship available at IBM Quantum.

  • Contact: Daniella Pombo, daniellapombo@ibm.com

AI Internship at CME Group

  • Position: Fall intern focusing on AI & Data.

  • Requirements: Strong background in CS/AI/ML, effective communication skills.

  • Contact: vijay.pillai@cmegroup.com

AI'S IMPACT ON VENTURE CAPITAL AND APPLICATION DEVELOPMENT

  • Insight from Byron Deeter, Bessemer Venture Partners:

    • AI is the major technological opportunity of our times, surpassing the cloud computing revolution.

    • Future of successful companies predicted to emerge in consumer-facing AI applications, with focus on productivity enhancement.

DATA MANAGEMENT AND ANALYTICS IN AI

  • Current Trends:

    • AI dominates headlines, yet demand remains for data professionals in data science and analytics fields.

    • Need for roles like data engineers and cloud data architects increases due to complexity of AI systems.

KEY FINDINGS FROM CISCO AI RESEARCH

  • Adoption Insights:

    • AI-ready companies outperform peers and are more likely to see success in innovation.

  • Barriers:

    • Rising workloads and lack of centralized data hinder AI adoption.

SCHEDULE AND TOPICS

  • Important Dates:

    • SKP Release: 10/23/25; Due: 10/30/25

    • Week 12 Topics:

    • Model evaluation for classification

    • Model selection and interpretability

    • Fairness in machine learning use cases

MEASURING ERROR OF REGRESSION MODELS

Assessing Model Accuracy

  • Objective: Assess how well model predictions match actual data.

  • Key Metrics:

    1. R² Statistic: Indicates the proportion of variance explained by the model.

    2. Mean Absolute Error (MAE): Average magnitude of errors in predictions.

    3. Mean Squared Error (MSE): Average of squared differences between predicted and actual values.

    4. Root Mean Squared Error (RMSE): Square root of MSE, providing error metric in same units as original data.

Error Function

  • Definition: Compares predictions with actual outcomes, returning smaller values for better predictions and larger values for poorer predictions.

REGRESSION MODEL FORMULA

  • Assumed model relationship: Y = f(X) + oldsymbol{oldsymbol{ ext{ϵ}}}

  • Linear approximation: Y = eta0 + eta1X + oldsymbol{oldsymbol{ϵ}}

MEASURING ACCURACY FOR CLASSIFICATION MODELS

  • Key Considerations:

    • Evaluation of prediction correctness is essential for model selection and feature refinement.

Classification Metrics

  • Accuracy: Determines overall correctness of predictions.

  • Precision: extPrecision=racTPTP+FPext{Precision} = rac{TP}{TP + FP}, focusing on minimizing false positives.

  • Recall (Sensitivity): extRecall=racTPTP+FNext{Recall} = rac{TP}{TP + FN}, focusing on minimizing false negatives.

  • F1-Score: F1=2imesracextPrecisionimesextRecallextPrecision+extRecallF1 = 2 imes rac{ ext{Precision} imes ext{Recall}}{ ext{Precision} + ext{Recall}}, achieving balance between precision and recall.

  • ROC and AUC: Measure aggregate performance across different classification thresholds. The AUC indicates overall classifier performance, with values closer to 1 indicating better performance.

CONFUSION MATRIX

  • Definition: Visual representation of prediction performance broken down by True Positives (TP), False Positives (FP), True Negatives (TN), and False Negatives (FN).

    • Key Metrics Derived:

    • Accuracy: (TP+TN)/(TP+FN+TN+FP)(TP + TN) / (TP + FN +TN + FP)

    • Precision: $TP / (TP + FP)$

    • Recall: $TP / (TP + FN)$

    • F1 Score: 2imes(PrecisionimesRecall)/(Precision+Recall)2 imes (Precision imes Recall) / (Precision + Recall)

  • Interpretation:

    • Diagonal elements reflect correct predictions, while off-diagonal elements reflect classification errors.

ERROR IN CLASSIFICATION MODELS

  • False Positive (FP): Incorrectly predicting true outcome as positive.

  • False Negative (FN): Incorrectly predicting a negative outcome as positive.

  • Examples:

    • Cancer diagnosis: High FP leads to unnecessary anxiety.

    • Fraud detection: High FN leads to financial losses.

ACCURACY EXPLAINED

  • Accuracy Formula: extAccuracy=racTP+TNTP+TN+FP+FNext{Accuracy} = rac{TP + TN}{TP + TN + FP + FN}

  • Concerns with Imbalanced Classes: Accuracy is insufficient; should accompany other metrics for robust evaluation.

METRICS TRADE-OFF AND INTERPRETATION

  • Interpreting Metrics:

    • High Precision & Low Recall: Model identifies few positives reliably.

    • High Recall & Low Precision: Model identifies most positives but includes negatives.

    • F1-Score: Provides single metric balancing Precision and Recall.

CLASSIFICATION USE CASE: MEDICAL DIAGNOSIS

Key Applications

  • Cancer detection from imaging (mammography).

  • Cardiac risk prediction from lab results.

  • Neurological disorder identification from clinical data.

Algorithms

  1. Convolutional Neural Networks (CNN)

  2. Random Forest

  3. Support Vector Machines (SVM)

  4. XGBoost

Challenges

  • Data imbalance in rare disease diagnosis.

  • Need for interpretability in clinical applications.

MODEL SELECTION IN MACHINE LEARNING

Overview

  • Purpose: Selecting the best model for task performance from candidates.

  • Stages:

    • Comparing algorithms (e.g., decision trees vs. neural networks).

    • Evaluating hyperparameters for model optimization.

Factors to Consider

  • Model training duration and explainability.

  • Impact of data preparation steps, pipeline considerations, and maintainability.

ETHICS AND FAIRNESS IN AI

  • Importance: Models must aim for fairness and avoid discrimination based on sensitive attributes (race, gender, etc.).

  • Legal & Ethical Obligations: Comply with regulations for fairness, transparency, and accountability.

  • Bias Sources: Unfair models emerge from biased data, inappropriate feature selections, and black-box nature of complex models.

LEGISLATIVE REGULATIONS

  • Recent regulations emphasize data protection, algorithmic transparency, fairness, and ethical AI use.

PRODUCT MANAGEMENT IN AI CONTEXT

Basket Composition & Product Substitution Models

  1. Model 1: K-means clustering for product basket analysis to ensure co-stocking.

  2. Model 2: Predictive modeling for substitution patterns using XGBoost regression.

PERFORMANCE METRICS - CASE STUDY**

Insights on Retail Impact

  • Demand patterns affected by COVID-19 and holiday peaks; elevated OOS rates observed.

  • Operational excellence achieved notable stock-out reductions.

Key Metrics

  • Customer satisfaction increases from improved product availability.

KEY TAKEAWAYS FOR AI PROJECTS

  1. Establish clear objectives and metrics before model development.

  2. Invest in A/B testing and experimental design to validate performance.

  3. Thoroughly explore data for informed modeling decisions.

ADDITIONAL RESOURCES

  • Investigate readings about deep learning and large language models.

  • Review ethical implications and regulatory responsibilities in AI deployment.

CLOSING REMARKS

  • For Q&A: Steven Keith Platt, Director of Analytics and Executive Lecturer of Applied AI, Loyola University Chicago.

  • Contact: splatt1@luc.edu