CSEG3060 · Research Methodology in Computer Science
Designing the talk, using audio-visual aids, communicating, handling queries — and the rubric you are scored against
Presentation is not a soft skill peripheral to research. It is a scientific act — scholarly argumentation where evidence, structure, rhetoric and visual design converge.
Unit IV · Lecture 3 of 3 · 60 minutes · CO4 / CO5 · B.Tech (CSE) Semester V
Dr. Mohsin Furkh Dar · School of Computer Science, UPES
→ / Space next · ← back · O overview · N speaker notes · F full screen
Learning outcomes · CO4 / CO5
Design a research presentation that aligns with academic review standards and fits its time budget
Deploy audio-visual aids that reinforce, rather than replace, oral argument
Demonstrate effective verbal and non-verbal communication during a scholarly defence
Anticipate, classify and answer questions raised by review committee members
Decode the evaluation rubrics committees use — and align the talk to the criteria that carry weight
Orientation
Structural design of the presentation — slide sequence, time allocation, cognitive load.
Multimodal communication: slides, demonstrations, figures — and their design rules.
Verbal, vocal and non-verbal delivery; the opening, the closing, and stage fright.
Strategic alignment with the evaluation criteria that the committee actually scores.
Rubric — a scoring framework of explicit criteria, performance descriptors and weightage.
Elevator pitch — a 60–90 second distillation of problem, motivation and expected contribution, used to open.
Back-pocket slide — a reserve slide answering an anticipated detailed query without disrupting the flow.
The guidance here follows standard practice in the IEEE, ACM and AAAI research communities, and peer-reviewed communication research — Mayer on cognitive load, Tufte on graphical integrity.
Pillar A · 1 of 3
Maximum, for a 20-minute talk. For a typical 15-minute B.Tech project review, 8–10 slides is optimal.
Maximum speaking time. The constraint is the audience’s attention, not your material.
Minimum size — legible from the back row of any standard lecture hall.
Each slide communicates one dominant cognitive unit. Mayer’s cognitive load theory of multimedia learning (2009) holds that working memory processes roughly 4±1 chunks simultaneously; overloading a slide measurably reduces comprehension.
Committees read an overrun as lack of rehearsal, poor understanding of your own work, or disrespect for their schedule — usually all three.
The safe strategy: rehearse to 80 percent of the allotted time, leaving 20 percent of buffer for the natural expansion that happens under questioning.
Pillar A · 2 of 3
| # | Content | Time | Strategic purpose |
|---|---|---|---|
| 1 | Title, authors, affiliation, guide, date | 30 sec | Establishes identity and context |
| 2 | Elevator pitch / problem statement | 90 sec | Hooks the committee and frames urgency |
| 3 | Motivation and background | 90 sec | Establishes the why now |
| 4 | Literature gap and research question(s) | 2 min | Demonstrates awareness of the state of the art |
| 5 | Objectives / hypothesis | 1 min | Crystallises the contribution claim |
| 6 | Proposed methodology | 3–4 min | The intellectual core of the talk |
| 7 | Expected outcomes / preliminary results | 2 min | Demonstrates progress and feasibility |
| 8 | Work plan and timeline (Gantt) | 1 min | Establishes realism of execution |
| 9 | References — five or six key works | 30 sec | Signals scholarship |
| 10 | Acknowledgements and Q&A invitation | 30 sec | Closes gracefully |
Pillar B · 1 of 4
Slides (PowerPoint, Beamer, Google Slides) · posters · printed handouts
Animations and transitions · live system demonstration · video clips · code walk-throughs in an IDE
Spoken narration · audio clips · music (rare, context-specific)
Interactive simulations · live polling (Mentimeter, Slido) · AR/VR demonstrations
Beamer (LaTeX) — publication-quality, ideal for theory-heavy talks · PowerPoint / Keynote — versatile, design-rich · Google Slides — collaborative · Overleaf + Beamer — manuscript and slides together · Marp, RevealJS — Markdown-based · OBS Studio, Loom — pre-recording · draw.io, Lucidchart — architecture diagrams · Matplotlib, Seaborn, Plotly — figures consistent with the proposal document
Pillar B · 2 of 4
High contrast between text and background — dark on light, or light on dark. Nothing in between.
Consistent fonts, colours and layouts across every slide.
Every element visually connected to at least one other element.
Group related items spatially — closeness is the claim of relationship.
Title 28–32 pt, bold · body 22–24 pt regular · captions and footnotes not below 18 pt.
Prefer sans-serif families — Calibri, Arial, Helvetica, Lato. They render better than serifs at projection distance.
60% dominant — usually a white background or the institutional colour
30% secondary — headings and emphasis
10% accent — reserved for novel contributions and key results
Use colour-blind-friendly palettes — viridis, ColorBrewer — wherever graphs and charts appear.
Pillar B · 3 of 4
| Pitfall | Consequence | Mitigation |
|---|---|---|
| Reading verbatim from slides | Loss of engagement; the committee perceives lack of depth | Slides are prompts, not scripts |
| Excessive animation | Cognitive distraction | Animate only to reveal a stepwise process |
| Low contrast, small fonts | Members at the back cannot read | Test from the back row during rehearsal |
| Cluttered graphs | Loss of insight | Apply Tufte’s data-ink ratio: maximise information, minimise chartjunk |
| Entire paragraphs embedded | Information overload | Replace prose with structured bullets and visuals |
Pillar C · 1 of 3
Optimal rate 130–150 words per minute — deliberately slower than conversation.
Strategic pauses of 1.5–2 seconds after key claims let the committee absorb and reflect.
Minimise fillers — um, uh, like, you know. A silent pause is always preferable to a filled one.
CS terms are routinely mispronounced. Settle yours in rehearsal:
LaTeX (LAH-tek or LAY-tek) · GIF (soft G, contested) · Ubuntu (oo-BOON-too) · et al. · ibid. (IB-id) · viz. (namely, videlicet)
Monotone delivery measurably reduces retention. Use pitch to signal meaning:
Rising — curiosity, open questions
Falling — definitive claims, conclusions
Lowered — emphasis on a critical claim
Eye contact
Distribute gaze across all members; never fixate on one person or the screen
Posture
Upright, open stance; no crossed arms, no leaning on the podium
Gestures
Indicate transitions, enumerate points, point to slide regions
Movement
Confined but purposeful — do not pace
Expression
Confidence and enthusiasm; smile at the opening and in acknowledgements
Mehrabian’s model is frequently over-applied, but its useful core holds: a large share of meaning in face-to-face communication travels non-verbally.
Pillar C · 2 of 3
A provocative question
“What if a neural network could explain its own decisions to a radiologist in real time?”
A striking statistic
“Globally, 2.5 quintillion bytes of data are generated daily, yet 80% remains unstructured and unanalysed.”
A compelling narrative
A brief anecdote of a real-world failure that motivates the research.
A clear thesis
“This work proposes a federated learning framework that reduces communication overhead by 40% while preserving differential privacy guarantees.”
A weak closing undermines the strongest presentation. The closing must echo the opening.
“In summary, we have presented a lightweight intrusion detection system for IoT networks that achieves 96.4% accuracy with 60% lower computational cost than the state of the art. This has direct implications for smart-city deployments. We will extend this to edge-fog architectures in the next phase. We welcome your questions and feedback.”
Pillar C · 3 of 3
Reframe the defence as a scholarly conversation, not a performance. You are the person in the room who has read the most about this problem.
Rehearse repeatedly in increasingly realistic conditions — alone, then to a peer, then in the actual room with the actual projector.
Box breathing (4-4-4-4) and progressive muscle relaxation, used in the ten minutes before you are called.
Arrive early, test the equipment, walk the room. Most acute pre-talk anxiety is displaced logistical anxiety.
Research support is contested — but the psychological priming effect is real enough to be worth two minutes.
Avoid stimulants before the talk. Caffeine converts nervousness into a visible tremor and an accelerated speaking rate.
Pillar D · 1 of 3
| Type | Intent | Example | Strategy |
|---|---|---|---|
| Clarification | Confirm understanding | “Could you clarify what you mean by ‘embedding space’?” | Paraphrase the question, then give a concise definition |
| Probing | Test depth of understanding | “Why did you choose Adam over SGD for this model?” | Explain the trade-offs; admit limitations where they exist |
| Challenging | Test defensibility of claims | “Your baseline comparison is weak; have you considered X?” | Acknowledge validity; defend or concede gracefully |
| Speculative | Explore extensions | “Could this method generalise to time-series data?” | Distinguish present scope from future work; commit if plausible |
| Procedural | Clarify logistics | “What is your data acquisition timeline?” | Answer directly; refer to the work plan slide |
Pillar D · 2 of 3
Interpret — repeat or paraphrase to confirm understanding. “If I understand correctly, you are asking whether our approach handles concept drift…”
Declare — state your position clearly. “Yes, we have addressed this through…” or “That is a limitation we acknowledge…”
Evidence — support with data, citations or logical reasoning
Assess — evaluate the implications. “This means our model can…”
Link back — connect the answer to the broader proposal narrative
Anticipate five to seven likely technical questions and prepare supplementary slides with the relevant data, equations or diagrams. Keep them after the visible deck and bring them up only when needed. This demonstrates preparedness without disrupting the narrative flow — and a presenter who can produce a convergence proof sketch on request has effectively ended that line of questioning.
Pillar D · 3 of 3
The committee is testing integrity as much as knowledge.
Fabricating an answer is the single most damaging error a presenter can make. An honest “I do not know, and here is how I will find out” costs a fraction of what an invented one costs when the committee follows it up.
Pillar D · alignment
| Criterion | Typical weight | What committees look for |
|---|---|---|
| Methodology | 20–25% | Scientific rigour; appropriateness of methods; reproducibility |
| Novelty and originality | 15–20% | Clear articulation of what is new; differentiation from prior work |
| Problem definition and motivation | 10–15% | Clarity of the problem statement; significance to discipline or society |
| Literature review | 10–15% | Coverage of seminal and recent work; identification of gaps |
| Feasibility and work plan | 10–15% | Realistic timeline; access to resources; risk mitigation |
| Expected outcomes and impact | 10–15% | Clarity of deliverables; academic and societal value |
| Presentation quality | 5–10% | Clarity, visual design, time management, Q&A handling |
| Ethics and compliance | 5% | IRB / IEEE ethics, data privacy, plagiarism declarations |
Strategic alignment
Emphasis on methodology, learning outcomes and demonstration.
Emphasis on literature review, contribution and methodological soundness.
Emphasis on broader impact, budget justification, deliverables and societal benefit.
Emphasis on novelty, technical depth and engagement.
Before the talk, identify the two or three criteria that carry the most weight in this specific context, and design the opening to maximise them.
For a DST-SERB proposal, lead with societal impact and deliverables. For a thesis defence, lead with methodological rigour and literature positioning. For an industry-sponsored project, lead with commercial viability and scalability.
Mandatory corrections — must be addressed before approval (a methodological flaw, missing ethical clearance). These determine acceptance.
Suggested improvements — strengthen the proposal if addressed (an additional baseline comparison). These determine quality.
Appreciative remarks — validate the approach and signal confidence. These are strategic guidance for publication.
Practical example
“Privacy-Preserving Federated Learning for Healthcare IoT” — team of four, assessed against the rubric on slide 15.
1 Title — project, team, supervisor, department, date. Clean institutional template.
2 Problem — “Hospitals cannot share patient data under HIPAA/GDPR, yet AI models require large datasets. How can we train a robust diagnostic model without centralising data?” Visual: split screen of two hospitals with a no-data-sharing lock.
3 Background — healthcare data breaches 2018–2024 (graph); growth of FL publications (graph). One bullet: “FL eliminates central aggregation but introduces privacy, communication and heterogeneity challenges.”
4 Gap — FedAvg, FedProx, SCAFFOLD in three columns with their limitations. Venn diagram of the challenges none fully address.
5 Objectives — (1) a DP-aware FL framework; (2) reduce communication rounds by 40%; (3) validate on three healthcare benchmarks.
6 Methodology — architecture diagram (clients → secure aggregation → global server with DP noise). Equation θ_{t+1} = θ_t − η∇L_DP(θ_t; ξ_t). Five-step workflow.
7 Preliminary results — table of accuracy, communication rounds and privacy budget ε against baselines; the proposed method in a coloured cell.
8 Work plan — Gantt for the remaining four months: experimentation (Jan–Feb), deployment (Mar), thesis writing (Apr).
9 References — McMahan et al. 2017, Abadi et al. 2016, Kairouz et al. 2021.
10 Closing — “We present a privacy-aware FL framework achieving 93.2% accuracy with 42% fewer communication rounds at ε = 1.0. We thank Dr. X and the IoT Lab, and welcome your questions.”
BPS-1 detailed convergence proof sketch · BPS-2 privacy budget analysis under composition theorems · BPS-3 comparison with FedBN for non-IID data · BPS-4 hyperparameter sensitivity plots
Before you present
Consolidation · end of Unit IV
Mock review panel, 10 marks (CO2, CO4), faculty and peer scoring. Use the slide-by-slide worked example as your template and prepare at least three back-pocket slides. You will be scored against the rubric on slide 15.
Mayer, Multimedia Learning (2009) · Tufte, The Cognitive Style of PowerPoint (2006) · Reynolds, Presentation Zen · your own department’s review rubric — request it before the panel
End of Unit IV — next, Unit V: Nature and Process of Intellectual Property.