Skip to content
View SaribAzim's full-sized avatar
4 followers · 7 following

Achievements

Achievement: Pull SharkAchievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull SharkAchievement: YOLOAchievement: Quickdraw

Block or report SaribAzim

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
SaribAzim/README.md

Typing SVG


Research Projects

Quantization Fidelity Study

Repository

Systematic evaluation of neural network quantization methods, investigating the trade-off between model precision, quantization error, and predictive performance.

Key finding
Quantization fidelity depends on weight rounding, quantization granularity, and calibration data, not just bit-width.

voiceMonitor · INFERENCE Lab

Repository Lab Note

Core Engineer on an open source vocal fatigue monitoring library (INFERENCE Lab Engineering Fellowship, Cohort 01). Designed a vocal load impulse response model adapted from the Banister sports science framework, with EMA smoothing, acoustic feature extraction (jitter, shimmer, HNR, CPPS), and personal baseline calibration. Shipped to v1.1.0 with 93% test coverage and CI.

Core idea
Vocal fatigue can be modelled like training load: each speaking session adds strain that decays over time, giving a recovery estimate instead of a single snapshot.

Published as an Engineering Journal entry on inference-lab.org

Neural Network Calibration Study

Repository

Post-hoc calibration analysis of ResNet-50 on CIFAR-10. Measured systematic overconfidence using ECE and MCE. Applied temperature scaling, a single scalar T, to reduce ECE by 67% without retraining.

Key finding
High top-1 accuracy and good calibration are independent properties. Most models only optimise for one.

Reference: Guo et al. (2017), ICML

CNN vs ViT Interpretability Study

Repository

Comparative saliency analysis across ResNet-50, EfficientNet-B0, and ViT-B/16 on ImageNette. Implemented Grad-CAM for CNNs and Grad-CAM with a patch reshape transform for ViT.

Key finding
CNNs focus on local discriminative patches. ViT attends globally, even when all three models agree on the correct label.

Reference: Selvaraju et al. (2017), ICCV

Technical Stack

Python PyTorch NumPy Git TypeScript Linux


Connect

LinkedIn GitHub



Pinned Loading

  1. quantization-fidelity-study quantization-fidelity-study Public

    How much model capability is lost when an LLM is quantized, and how much of that loss can external knowledge retrieval recover without restoring the original model precision?

    Python

  2. Neural-Network-Calibration Neural-Network-Calibration Public

    Post-hoc calibration study using temperature scaling on CIFAR-10 — ECE reduced 67% with a single learned parameter. Based on Guo et al. (2017)

    Python

  3. Inference-LAB/voiceMonitor Inference-LAB/voiceMonitor Public

    Python 5

  4. SaadAhmed29/BioFusion-AI SaadAhmed29/BioFusion-AI Public

    Cross-modal breast cancer diagnostic tool combining CNNs and clinical risk factors to enhance radiologist decision-making. 92% accuracy with ensemble learning.

    Python

  5. CNN-VIT-Interpretability-Study-with-Grad-CAM-and-Reshape-Transform- CNN-VIT-Interpretability-Study-with-Grad-CAM-and-Reshape-Transform- Public

    Interpretability study comparing how ResNet-50, EfficientNet-B0, and ViT-B/16 visually reason about images — using Grad-CAM for CNNs and Grad-CAM with reshape transform for ViT, across 10 ImageNett…

    Python

  6. Attention-Based-BiLSTM-for-Multilingual-Fake-News-Detection Attention-Based-BiLSTM-for-Multilingual-Fake-News-Detection Public

    Python