1. Which application is LEAST suitable for approximate computing?
A. Real-time video game physics & rendering
B. Bank account balance calculations
C. Convolutional Neural Network (CNN) image classification
D. Audio noise reduction and speech recognition
2. Why can neural networks and human eyes tolerate approximate arithmetic?
A. Because pixels are always binary 0 or 1
B. Because digital multipliers are completely free
C. Because perception and deep learning weights have inherent error tolerance
D. Because exact math creates compiler errors
3. What causes the main speed bottleneck (critical path delay) in an exact binary adder?
A. Carry bit propagation having to ripple all the way from Bit 0 to Bit 15
B. The clock frequency being too high
C. Too many AND gates generating partial products
D. DRAM refresh cycles
4. How does the Rounding-Based Multiplier (RoBA) achieve over 70% energy savings?
A. By turning off the chip power supply
B. By rounding inputs to powers of two ($2^k$), converting multiplication into simple wire shifts
C. By calculating 64-bit Taylor expansions
D. By running the clock backwards
Completed:
0
/4 | Score:
0
/4
🔄 Retake Quiz