Master Thesis - Approximative Computing for CNNs

Master Thesis - Approximative Computing for CNNs

Category Computer Science, Machine Learning, Electronics, Integrated Circuits

Scope Two Master students completing 30 credits each

Background Convolutional neural networks for object detection and semantic segmentation are increasingly deployed directly on edge devices, where power consumption and silicon area are tightly constrained. Most of the energy and area in a neural network accelerator is spent on multiply-accumulate operations. Approximate computing offers a way to reduce these costs by replacing exact arithmetic circuits with simpler alternatives that trade numerical accuracy for lower power consumption and smaller circuit area. Because neural networks are inherently tolerant to numerical noise, approximate multipliers have shown promising results for image classification. However, their impact on dense-prediction tasks such as object detection and semantic segmentation, where precise localization and per-pixel accuracy matter, remains largely unexplored.

Goal This thesis will investigate how approximate arithmetic circuits affect the accuracy and hardware efficiency of CNNs for object detection and semantic segmentation. Using established open-source libraries of approximate multipliers and simulation frameworks, you will:

• Evaluate how approximate multipliers affect detection and segmentation performance.
• Quantify the trade-off between network accuracy and estimated power and area savings.
• Explore methods for recovering lost accuracy, such as layer-wise selection of multipliers, calibration, and fine-tuning.

Who are you? We're looking for two Master's students in Electrical Engineering, Computer engineering, Computer science, or a related field who enjoy working where hardware meets machine learning. You're curious about how algorithms map onto silicon and motivated by turning experimental results into clear, well-founded conclusions.

You should have:

• Solid programming skills in Python, and experience with a deep learning framework such as PyTorch or Tensorflow
• A basic understanding of convolutional neural networks and how they are trained and evaluated
• Coursework or experience in digital design or computer architecture, including computer arithmetic

It's a plus if you have:

• Experience with object detection or semantic segmentation models
• Familiarity with neural network quantization or hardware-aware model optimization
• Experience with hardware description languages (Verilog/VHDL) or synthesis tools
• Knowledge of C/C++ and/or CUDA
• An interest in energy-efficient computing and edge AI

If you're applying as a pair, we especially welcome complementary profiles, for example one person with a stronger machine learning background and one with a stronger hardware background.

Type of Employment Thesis Worker (Fixed Term)

Posting End Date 2026-10-14

Certain roles at Axis require background checks, which means applicable verifications will be done in these recruitments. Notice will be provided before we take any action.

About Axis Communications We enable a smarter and safer world by creating innovative solutions for improving security and business performance. As a network technology company and industry leader, we offer solutions in video surveillance, access control, intercom, and audio systems, enhanced by intelligent analytics applications.

With around 6,000 committed employees in over 50 countries, we collaborate with partners worldwide. Together, we thrive in our friendly, open, and collaborative culture and inspire each other to think beyond the expected. United by our commitment to inclusion, diversity, and sustainability, we consistently seek to develop our skills and way of working.

Let´s create a smarter and safer world

For more information about Axis, please visit our website www.axis.com.

Listen to Axis Communications Playlist