Research & Inquiry

Documenting experiments, discoveries, hypotheses & evolving intellectual frameworks.

This is where I explore technical questions, test assumptions, analyze systems, and share learnings from iterative experimentation.

Filter Research

Keywords

Year

Publication

Status

Showing 3 of 3 entries

Research Entries (3)

Inference Latency on Edge Devices
How much can inference speed be improved via quantization + operator fusion on Raspberry Pi?
1 min read

This research investigates methods to improve deep learning inference speed on resource-constrained edge devices. We explore quantization techniques (FP32β†’INT8), operator fusion, and compare ARM CPU versus VideoCore GPU acceleration paths. Results show 2.4x speedups with quantized models while maintaining accuracy within acceptable thresholds.

ongoing
August 15, 2024
edge computing
performance optimization
Noise Reduction in Sensor Fusion
To what extent can complementary filters outperform naive averaging for IMU stability?
1 min read

This study compares classical complementary filtering with moving average techniques for inertial measurement unit (IMU) data fusion. A complementary filter implementation in C++ on Arduino platforms demonstrates significant reduction in high-frequency jitter and gyroscope drift, with practical applications in robotics and drone navigation.

concluded
June 21, 2024
sensor fusion
signal processing
Small-Scale Personalization of Language Models
How do LoRA adapters affect internal representation geometry of embeddings?
1 min read

This research investigates the impact of parameter-efficient fine-tuning (LoRA adapters) on language model behavior and internal representations. We analyze whether adopting a conversational persona affects core reasoning capabilities and explore the geometry of embedding space changes during adaptation. Results show minor (4%) degradation on general reasoning while successfully adopting target personas.

in revision
March 10, 2024
large language models
fine-tuning

Learning Log

2024-06-21

Kalman filters require good system models; complementary filters work even with imperfect assumptions.

2024-08-17

Attention heads in transformers appear to self-specialize even without explicit constraints.