---
title: "Berkan Sesen (@berkansesen) | Hashnode"
description: "I think about AI. A lot. Read the latest articles by Berkan Sesen on Hashnode."
url: "https://hashnode.com/@berkansesen"
---

@berkansesen

# Berkan Sesen

@berkansesen · London, UK · Joined March 2026

I think about AI. A lot.

### [Handling Class Imbalance: Downsampling, SMOTE, and Beyond](https://sesenai.hashnode.dev/handling-class-imbalance-downsampling-smote-and-beyond?utm_source=hashnode&utm_medium=feed)

Compare five strategies for imbalanced classification: random downsampling, upsampling, SMOTE, and class weights. Evaluate with precision-recall curves and F1, not just accuracy.

13 min read · [sesenai.hashnode.dev](https://sesenai.hashnode.dev/handling-class-imbalance-downsampling-smote-and-beyond?utm_source=hashnode&utm_medium=feed)

[Jul 5](https://hashnode.com/posts/handling-class-imbalance-downsampling-smote-and-beyond/6a4a4b6560ef028b52111076)

### [AIC and BIC: Choosing the Right Model Without Overfitting](https://sesenai.hashnode.dev/aic-and-bic-choosing-the-right-model-without-overfitting?utm_source=hashnode&utm_medium=feed)

Derive the log-likelihood from first principles, then build AIC and BIC as penalised model selection criteria. Compare polynomial fits, visualise the bias-variance tradeoff, and understand when to use each criterion.

13 min read · [sesenai.hashnode.dev](https://sesenai.hashnode.dev/aic-and-bic-choosing-the-right-model-without-overfitting?utm_source=hashnode&utm_medium=feed)

[Jun 29](https://hashnode.com/posts/aic-and-bic-choosing-the-right-model-without-overfitting/6a422fe2328f4f8cb9f3575d)

### [Trust Region Methods: From REINFORCE to TRPO to PPO](https://sesenai.hashnode.dev/trust-region-methods-from-reinforce-to-trpo-to-ppo?utm_source=hashnode&utm_medium=feed)

Implement REINFORCE, TRPO, and PPO from scratch in PyTorch on CartPole. Understand why vanilla policy gradients are unstable, how trust regions fix it, and why PPO became the default algorithm in modern RL.

21 min read · [sesenai.hashnode.dev](https://sesenai.hashnode.dev/trust-region-methods-from-reinforce-to-trpo-to-ppo?utm_source=hashnode&utm_medium=feed)

[Jun 17](https://hashnode.com/posts/trust-region-methods-from-reinforce-to-trpo-to-ppo/6a325fcda74418b00044d612)

### [LDA vs PCA: Supervised Meets Unsupervised Dimensionality Reduction](https://sesenai.hashnode.dev/lda-vs-pca-supervised-meets-unsupervised-dimensionality-reduction?utm_source=hashnode&utm_medium=feed)

Compare PCA and Fisher's Linear Discriminant Analysis on the Iris dataset. Learn when unsupervised reduction fails, why labels matter, and the eigenvalue problem behind both methods.

13 min read · [sesenai.hashnode.dev](https://sesenai.hashnode.dev/lda-vs-pca-supervised-meets-unsupervised-dimensionality-reduction?utm_source=hashnode&utm_medium=feed)

[Jun 12](https://hashnode.com/posts/lda-vs-pca-supervised-meets-unsupervised-dimensionality-reduction/6a2c3271d617a213c25a71ce)

### [Changepoint Detection: Finding Regime Shifts in Financial Data](https://sesenai.hashnode.dev/changepoint-detection-finding-regime-shifts-in-financial-data?utm_source=hashnode&utm_medium=feed)

Detect volatility regime shifts in the S&P 500 using the PELT algorithm in Python. Compare PELT with Binary Segmentation, tune the penalty parameter, and align detected changepoints with real financial crises.

14 min read · [sesenai.hashnode.dev](https://sesenai.hashnode.dev/changepoint-detection-finding-regime-shifts-in-financial-data?utm_source=hashnode&utm_medium=feed)

[Jun 9](https://hashnode.com/posts/changepoint-detection-finding-regime-shifts-in-financial-data/6a28269fa739a2faab98232c)