Background and Professional Experience

• Interdisciplinary Perspective in Education

  • Combines technical marketing, computer science, and mathematics.
  • Focuses on enriching STEM education and enhancing student learning.

• Industry Background

  • Technical Marketing Engineer at Intel Corporation (Rio Rancho, NM)
  • Collaborated with product divisions to develop software ecosystems for new platform features.
  • Led a diverse team of marketing engineers, showcasing emerging technologies.
  • Appreciated for bridging the gap between technical communication and strategic marketing.

• Academic Leadership

  • Former CIS Program Chair at Central New Mexico Community College.
  • Taught various courses such as XML, project management, and Microsoft Office tools.
  • Positions held at San Juan College and Pikes Peak State College demonstrated commitment to student success.

• Academic Qualifications

  • Master of Science in Mathematics from New Mexico Tech.
  • Thesis on feedforward neural networks.
  • Doctoral coursework in statistics at the University of New Mexico.
  • Bachelor’s degrees:
  • Bachelor of Arts in Psychology from Murdoch University (WA, Australia).
  • Bachelor of Science in Mathematics (Summa Cum Laude) from Southern Oregon State.
  • Industry certifications include MCSE, MCP Visual Basic, and MOS Access, showcasing continuous professional development.

Neural Networks & Artificial Intelligence

• Introduction to AI

  • Artificial Intelligence (AI) began to develop in the late 1950s.
  • Early AI models like the perceptron were foundational, though limited by linear separability.

• Perceptron Model

  • Concept:
  • Inspired by biological neurons; processes inputs to classify between two classes (binary classification).
  • Components:
  • Dendrites: Receive inputs (features).
  • Cell body: Processes inputs and weights.
  • Axon: Outputs decision after applying activation function.
  • Bias: A constant that aids decision-making.
  • Learning Process:
  • Utilizes supervised learning; learns through labeled data.
  • Error calculation method involves the true label vs. predicted output.
  • Adjustments of weights and bias based on errors help refine the model.

• Limitations of Basic Perceptrons

  • Linear Separability:
  • Successful at classifying linearly separable data but fails with non-linear problems (e.g., XOR problem).
  • This limitation resulted in the AI winter period (1969-1992).

• Backpropagation and Revivals

  • Introduction of backpropagation helped solve more complex problems by allowing multilayer networks.
  • Architecture: Multiple hidden layers process inputs for non-linear combinations, vastly increasing capability.
  • Gradient Descent: Method to minimize errors through adjusting weights and biases during the learning process.

• Deep Learning Era

  • Modern Advances:
  • Proliferation of deep learning techniques, especially post-2012.
  • Modern Neural Networks:
  • Capable of processing vast amounts of data (e.g., images, text) through layered networks.
  • Marked improvement in performance for tasks like image and speech recognition.

• Transformer Architecture

  • Adopted for natural language processing; enhances learning from complex text data.
  • Maps textual data into vectors, enabling nuanced understanding and generation.
  • Components:
  • Multi-head self-attention: Assesses word importance across contexts.
  • Feed-forward networks: Similar to basic perceptrons but applied in a more complex structure.

• AI Limitations and Concerns

  • Known phenomena such as "hallucination" (AI creating plausible but incorrect information).
  • Importance of verifying AI responses against trusted sources for accuracy.

• Comparative Scale

  • Current AI models (e.g., ChatGPT, Grok) operate with trillions of parameters compared to earlier versions (thousands to millions).
  • Significantly better data handling due to technological advancements (e.g., GPUs, TPUs).
  • Limitations in Understanding AI Output:
  • AI systems are often termed black boxes; their decision-making processes can be opaque.

Conclusion

• Professor Jonas Morrison exemplifies an interdisciplinary and innovative educator.
• Neural networks have evolved from simple binary classification models to complex deep learning systems capable of solving numerous real-world challenges, albeit with certain constraints.
• Future advancements in AI will likely demand continued attention to ethics, accuracy, and data accountability.