--- title: "VACSEN: A Visualization Approach for Noise Awareness in Quantum Computing" authors: - Shaolun Ruan - Yong Wang - Weiwen Jiang - Ying Mao - Qiang Guan abstract: "Quantum computing has attracted considerable public attention due to its exponential speedup over classical computing. Despite its advantages, today’s quantum computers intrinsically suffer from noise and are error-prone. To guarantee the high fidelity of the execution result of a quantum algorithm, it is crucial to inform users of the noises of the used quantum computer and the compiled physical circuits. However, an intuitive and systematic way to make users aware of the quantum computing noise is still missing. In this paper, we fill the gap by proposing a novel visualization approach to achieve noise-aware quantum computing. It provides a holistic picture of the noise of quantum computing through multiple interactively coordinated views: a Computer Evolution View with a circuit-like design overviews the temporal evolution of the noises of different quantum computers, a Circuit Filtering View facilitates quick filtering of multiple compiled physical circuits for the same quantum algorithm, and a Circuit Comparison View with a coupled bar chart enables detailed comparison of the filtered compiled circuits. We extensively evaluate the performance of VACSEN through two case studies on quantum algorithms of different scales and in-depth interviews with 12 quantum computing users. The results demonstrate the effectiveness and usability of VACSEN in achieving noise-aware quantum computing." summaryType: survey sourceStatus: null sources: - https://doi.org/10.1109/TVCG.2022.3209455 --- # VACSEN: A Visualization Approach for Noise Awareness in Quantum Computing [Original paper](https://doi.org/10.1109/TVCG.2022.3209455) ## Background and motivation VACSEN helps quantum computing users select a quantum computer and a compiled physical circuit by inspecting their noise characteristics before execution. A logical circuit can be mapped to different physical qubits and gates, each with different reliability, and the same hardware's calibration properties change over time. Consequently, choosing a backend with sufficient qubits and choosing a shallow compiled circuit do not by themselves resolve the execution-quality problem. Cloud queues add another consideration: users may wait a long time for a run whose result is subsequently too noisy to be useful. The paper situates this workflow in the noisy intermediate-scale quantum computing platforms available during its development and evaluation in 2022. Noise and hardware-aware circuit mapping are established problems. Related work improves qubit hardware, applies quantum error correction, or optimizes mapping through qubit reliability, connectivity constraints, and gate reordering. The authors argue that these approaches still leave users needing to understand the noise of available computers and alternative compiled circuits. They also distinguish VACSEN from quantum visualization tools such as GraphStateVis, QuFlow, Quirk, and vendor circuit interfaces, whose emphasis is on circuits, states, or algorithm behavior. Vendor calibration indicators expose individual hardware properties, but the paper identifies a gap in coordinated visual analysis connecting temporal hardware behavior to component usage in a particular compiled circuit. The contribution therefore addresses an existing reliability problem through a new visual analysis workflow; it does not introduce noise itself as a new problem or establish a new error-correction method. Figure 2 motivates this distinction with a two-qubit circuit mapped onto different adjacent qubit pairs of the five-qubit `ibmq_bogota` device. Its circuit and mapping diagrams accompany measured output probabilities from January 8, 2022, showing that different placements can produce different results even without inserted SWAP gates. This is a motivating hardware example, separate from the VACSEN interface and its subsequent evaluation. ## Design study and system workflow The design was developed with five quantum computing experts over five months. The pilot involved interviews, an initial prototype, and iterative feedback, including weekly meetings with two experts experienced in temporal hardware-quality analysis. One of the five pilot participants was a coauthor. The resulting six requirements cover temporal noise analysis, access to the latest available noise and queue information, an overview of compiled circuits, detailed comparison of qubit and gate usage, compilation and execution for fidelity validation, and flexible interactions with understandable encodings. The implemented workflow connects IBM Quantum calibration and execution data to three main coordinated views. As shown schematically in Figure 3, a storage module maintains qubit and gate properties and their reference values, a processing module handles data preparation and circuit compilation or extraction, and a visualization module supports analysis. The collected properties include relaxation time $T_1$, dephasing time $T_2$, readout error, gate error rate, queue length, and the usage counts of qubits and gates in compiled circuits. Users select a computer, compile an algorithm repeatedly to obtain candidate mappings, filter those candidates, compare component usage, and submit a selected circuit for execution. Execution results support a subsequent fidelity comparison, which the paper calculates using Hellinger distance between ideal and observed output distributions. Figure 1 shows the implemented interface during the Shor circuit case study. Panels A through C link computer selection to candidate-circuit filtering and detailed comparison; D contains the controls, while E and F display execution results. The thin connections between views associate selected computers and circuits across these levels of analysis. ## Visual encodings and interactions ### Temporal comparison of quantum computers The Computer Evolution View represents each computer as a sequence of circuit-like time slices. Each slice aligns qubits vertically and depicts one user-selected qubit metric with circles. Circle color indicates performance relative to a mean reference value: blue means better, while red means worse. Radius encodes the magnitude of the difference, so a large circle is an extreme value rather than necessarily a good value. For readout error, smaller errors are better; for $T_1$ and $T_2$, longer times are better. This relative encoding emphasizes differences across the available hardware instead of requiring users to compare raw numbers throughout the display. Gate connections appear as gray line segments whose endpoints identify the two connected qubits. Within a time slice, a segment's horizontal position represents gate error rate, allowing topology and gate quality to be read together. Opacity reduces the visual impact of overlapping segments. A Gaussian kernel density estimate below each slice summarizes the distribution of gate errors, with a blue-to-red gradient indicating increasing error. The sequence of slices exposes changes over time, while horizontal bars next to computer names show current queue lengths. Users can change the displayed time range, sampling interval, and qubit metric. Figure 4 explains the final encoding and the alternatives considered during design. The alternatives include separate temporal curves, highlighted anomalous intervals, a layout connecting qubit and gate rows, and a time-slice layout without the final gate-error distribution encoding. The authors report that these alternatives either obscured trends in limited space, omitted connectivity, introduced clutter, or communicated gate noise less clearly. These are design-study rationales based on expert feedback, rather than results of a controlled comparison proving one encoding universally superior. ### Filtering candidate compiled circuits The Circuit Filtering View allocates one row to each compiled circuit. A circle's horizontal position represents circuit depth, while its color and radius represent whether a selected performance score is above or below the mean and by how much. Blue denotes a higher score, and red a lower score. Users switch between gate and qubit scores, sort by score or depth, restrict the score range with a slider, and select candidates for detailed inspection. The control panel also selects an algorithm and the number of compilation attempts. The paper explicitly defines the performance score as the reciprocal of the usage-weighted mean error rate: $$ S = \left(\frac{\sum_{i=1}^{N} C_i E_i}{\sum_{i=1}^{N} C_i}\right)^{-1}, $$ where $E_i$ is a physical gate's or qubit's error rate and $C_i$ is its usage count in the compiled circuit. This makes frequently used components contribute more strongly than rarely used ones and assigns larger scores to smaller weighted error rates. The score provides a compact screening measure; it is distinct from the fidelity calculated after execution, and the paper does not establish it as an exact prediction of end-to-end circuit fidelity. ### Detailed circuit comparison and execution results Each selected circuit is represented by a coupled bar chart. The upper bars encode gate usage counts, and the lower bars encode qubit usage counts. Their heights represent usage, while color represents gate error or the selected qubit metric, using blue for better performance and red for worse performance. Black outlined reference rectangles show each component's mean usage across the compiled circuits. Curved lines connect each gate to the two qubits it operates on. Users can therefore identify circuits that repeatedly use a particularly noisy component, even when their aggregate scores or depths appear similar. This display summarizes component usage and connectivity rather than preserving the complete temporal ordering of circuit instructions. Figure 5 contrasts the coupled bar chart with a topology layout using circle sizes and edge widths for usage, and another design mixing vertical gate bars with horizontal qubit positions. The final design follows concerns about comparing edge widths and switching between different comparison directions. After execution, the Fidelity Comparison View positions a mark vertically by the circuit's fidelity and highlights the selected circuit in black. Selecting a result reveals its output distribution in the Probability Distribution View, where basis-state outcomes lie on the horizontal axis and shot counts on the vertical axis. Figure 1F places noise-free and experimental counts side by side, connecting the calibration-based analysis to observable execution results. ## Evaluation and findings Two case studies used IBM Quantum on March 5, 2022, with two experts who also participated in the subsequent interview study. In the two-qubit case, the user compared a week of hardware behavior and queue lengths, choosing `ibmq_manila` instead of the more heavily queued `ibm_perth`. Sixty compilation attempts produced four mapping categories with the same reported depth of 24. The user selected `trans_24` after inspecting its gate and qubit quality; its reported fidelity was 89.4%, compared with 87.5%, 83%, and 82.1% for the other three tested mappings. A further selection on the noisier but lightly queued `ibmq_lima` achieved 83.1%, compared with 77.5%, 74.4%, and 58.5% for its alternatives. Figure 6 documents the temporal hardware views, candidate mappings, coupled bars, and resulting fidelity comparisons for this case. The second case used a seven-qubit implementation of Shor's algorithm. The user compared hardware using readout error, gate error, $T_1$, and queue information, then compiled 60 candidates on `ibmq_jakarta` with reported depths from 706 to 817. After selecting the five shallowest candidates, the coupled bar charts revealed differences in repeated use of a noisy gate and a noisy qubit. The selected circuit, `trans_19`, achieved the highest fidelity among those five at 73.6%. The paper reports that it and `trans_40` formed the higher-fidelity group, while the other three averaged 58%. These cases illustrate useful selection reasoning and favorable outcomes among the inspected alternatives; they do not establish that VACSEN always finds a globally optimal mapping. The interview study recruited 12 quantum computing participants distinct from the pilot experts, spanning postgraduate students, research staff, and professors; none had a visualization or HCI background. Participants chose among QFT, Bernstein–Vazirani, and Shor circuits, received a roughly 25-minute tutorial, completed nine analysis tasks for about 40 minutes while thinking aloud, and took part in a roughly 25-minute post-study interview. They also answered 12 seven-point Likert questions. The reported mean ratings were 5.83 for noise-awareness effectiveness, 5.88 for usability, 6.02 for visual design, and 6.04 for interactions. Figure 7 presents response distributions grouped by those four aspects. Participants valued temporal comparison, linked hardware and circuit analysis, and integration with remote execution. The evidence supports perceived usefulness and the demonstrated workflows, but the study does not provide a controlled baseline comparison of task accuracy, completion time, or fidelity improvement across users. ## Contributions, limitations, and future work The main contributions are the expert-informed requirements, an implemented workflow linking hardware history to compiled-circuit selection and execution, and the circuit-like temporal view and coupled bar chart that make this analysis possible. The paper's broader design lesson is that quantum computing users benefited from selecting among familiar, manageable encodings rather than seeing every noise attribute simultaneously. VACSEN provides information for human selection; automatic recommendations remain future work. Calibration availability constrains the system's claim to show the latest noise information. The authors observed update intervals longer than a day and pauses lasting several days, so retrieving the latest available calibration does not provide continuously refreshed physical noise measurements. They also acknowledge visual scalability limitations for much larger devices, giving 127 qubits as an example beyond the small-device settings emphasized in their evaluation. The paper does not identify a tested failure threshold at 50 qubits. The implementation was mainly tested with IBM Quantum data. Extension to other platforms is proposed on the basis of similar workflows and performance attributes, rather than demonstrated through a cross-platform evaluation. Future work includes improving scalability and automatically recommending computers and compiled circuits. Participants additionally suggested considering correlations among noise types, supporting more efficient execution, and extending the system toward quantum machine learning workflows.