AI.Cruz - Learn AI Through Interactive Visualizations

RESEARCH

Conferences, Venues & Peer Review

NeurIPS, ICML, and ICLR are the top ML venues. Understanding acceptance rates, review timelines, and how to respond to reviews can double your publication rate.

1Top AI/ML Venues

In ML, conferences are primary publication venues — not journals. The field moves too fast for journal timelines. A paper at NeurIPS 2024 will be read by thousands within days of acceptance; the equivalent journal paper might take 18 months to appear. Understanding the venue hierarchy helps you target your work appropriately.

NeurIPS
Neural Information Processing Systems. Flagship ML/AI conference. ~14,000 submissions, ~25% acceptance (2024). Held in December. Accepts ML theory, systems, applications, and neuroscience.

ICML
International Conference on Machine Learning. ~12,000 submissions, ~28% acceptance (2024). Held in July. Slightly more theory-focused than NeurIPS. Both are top-1 venues in industry and academia.

ICLR
International Conference on Learning Representations. ~9,000 submissions, ~31% acceptance (2024). Held in May. Unique: reviews are public on OpenReview, making it the most transparent venue.

CVPR / ECCV / ICCV
Computer Vision top venues. CVPR is the largest (~25k submissions). ECCV and ICCV alternate years. Accept rates ~20-25%. If your work is primarily vision, target these over NeurIPS.

ACL / EMNLP / NAACL
Natural Language Processing venues. ACL is the flagship. All three now use ARR (ACL Rolling Review) — submit once, get reviews, choose which venue to commit to. Better than repeated re-submission.

Workshops
Attached to all major conferences. Lower bar, faster turnaround. Good for preliminary work. Workshop papers are not counted as full publications in most CVs, but the feedback is invaluable.

2The Review Process

All major ML venues use double-blind peer review: authors do not know who the reviewers are, and reviewers do not know who the authors are. Each paper receives 3-4 reviews. An area chair (AC) reads the paper and reviews, mediates disagreements, and makes the final recommendation to the programme chairs.

Typical ICLR timeline: submission deadline (October) → reviews released (November) → author rebuttal window (2 weeks) → reviewer discussion (December) → decisions (January) → camera-ready (February) → conference (May). The whole cycle is ~7 months.

What reviewers look for

✓ Significance and novelty of contribution
✓ Correctness of claims and experiments
✓ Comparison to relevant baselines
✓ Clarity of presentation

Common rejection reasons

✗ Incremental contribution over prior work
✗ Missing key baselines
✗ Inconsistent or cherry-picked results
✗ Insufficient ablations

3Writing and Responding to Reviews

The rebuttal is your one opportunity to address reviewer concerns before the final decision. Most authors make the same mistake: they argue against the reviewers instead of addressing their concerns. A good rebuttal is a promise — "we will add X experiment" — not a debate.

Rebuttal principles: address every reviewer point, even those you disagree with. Be polite and professional — reviewers are volunteers. Provide concrete additions ("we added Table 3 in the supplementary material"). Do not add large amounts of new content. Match the reviewer's framing. A reviewer who feels heard is more likely to raise their score.

One experiment per concern
Add at most one new result per major reviewer concern. More than that looks like you are papering over weaknesses rather than addressing them.

Acknowledge weaknesses
If a reviewer identifies a real limitation, acknowledge it and explain why it does not invalidate the contribution. Denial makes reviewers more suspicious, not less.

Rejection is data
Every rejection with reviews is feedback. Top researchers average 1-2 rejections per accepted paper. Treat reviews as a free pre-submission to the next venue.

4Preprints, arXiv, and Open Access

The ML community has effectively moved to open-access publishing. Nearly every paper appears on arXiv — often the day before, day of, or day after conference submission. arXiv preprints are indexed by all major search engines, citable, and freely accessible to anyone. There is no longer a paywall between the public and cutting-edge ML research.

arXiv preprints
Submit to arXiv (cs.LG, cs.AI) before or after conference submission. Establishes a public timestamp. Most venues explicitly permit preprints under double-blind review.

ICLR open review
All ICLR reviews are public on OpenReview — accepted and rejected alike. This is a major cultural difference from NeurIPS/ICML. Rejected ICLR papers with public reviews can still circulate.

Hugging Face visibility
Hugging Face Daily Papers surfaces the most-upvoted arXiv submissions each day. Getting your paper on that list dramatically increases initial readership and citations in the first weeks.

Track conference deadlines and days remaining

from datetime import date, timedelta

CONFERENCES = [
    {"name": "NeurIPS 2025",    "deadline": date(2025, 5, 22),  "venue": "Vancouver",    "tier": 1},
    {"name": "ICML 2025",       "deadline": date(2025, 1, 23),  "venue": "Vienna",       "tier": 1},
    {"name": "ICLR 2025",       "deadline": date(2024, 10, 7),  "venue": "Singapore",    "tier": 1},
    {"name": "CVPR 2025",       "deadline": date(2024, 11, 15), "venue": "Nashville",    "tier": 1},
    {"name": "EMNLP 2025",      "deadline": date(2025, 6, 1),   "venue": "Honolulu",     "tier": 2},
    {"name": "NeurIPS 2025 WS", "deadline": date(2025, 9, 5),   "venue": "Vancouver",    "tier": 3},
]

def deadline_status(conf: dict, today: date) -> str:
    days = (conf["deadline"] - today).days
    if days < 0:
        return f"CLOSED ({abs(days)} days ago)"
    if days == 0:
        return "TODAY"
    if days <= 7:
        return f"URGENT ({days} days)"
    if days <= 30:
        return f"SOON ({days} days)"
    return f"{days} days"

def progress_bar(deadline: date, today: date, width: int = 20) -> str:
    total = 180  # assume 6-month window
    elapsed = max(0, min(total, total - (deadline - today).days))
    filled = int(width * elapsed / total)
    bar = "#" * filled + "-" * (width - filled)
    return "[" + bar + "]"

today = date.today()
tier_labels = {1: "Tier-1", 2: "Tier-2", 3: "Workshop"}

print(f"Conference Deadline Tracker — {today}")
print("=" * 65)

for tier in [1, 2, 3]:
    confs = [c for c in CONFERENCES if c["tier"] == tier]
    if not confs:
        continue
    print(f"\n{tier_labels[tier]}")
    for c in sorted(confs, key=lambda x: x["deadline"]):
        status = deadline_status(c, today)
        bar = progress_bar(c["deadline"], today)
        print(f"  {c['name']:<20} {c['deadline'].strftime('%b %d')}  {bar}  {status}")

print()
print("Tip: a paper rejected from NeurIPS (May) can be resubmitted to ICLR (Oct)")
print("with reviewer feedback incorporated — a 4-month turnaround.")

# Calculate ICLR -> NeurIPS revision window
try:
    iclr = next(c for c in CONFERENCES if "ICLR" in c["name"])
    neurips = next(c for c in CONFERENCES if "NeurIPS 2025" == c["name"] and "WS" not in c["name"])
    gap = neurips["deadline"] - iclr["deadline"]
    print(f"\nGap ICLR → NeurIPS deadline: {gap.days} days ({gap.days // 30} months)")
except StopIteration:
    pass