Scientify
Open workspace

Complete an entire
research project in the cloudFrom one research question to a report, code and data.

Open workspace See an example task
Research task · Example
  1. Search papers
  2. Read LOAD_DATA
  3. Run RUN_EXPERIMENT
  4. Write ERROR_COMPARISON
Review the last three years of papersReproduce a paper's experimentCompare forecast errors across modelsRun a survival analysisBuild an epidemic transmission modelSolve a math modeling problemTest the difference between two groupsClean raw dataFind research gaps
Review the last three years of papersReproduce a paper's experimentCompare forecast errors across modelsRun a survival analysisBuild an epidemic transmission modelSolve a math modeling problemTest the difference between two groupsClean raw dataFind research gaps
Review the last three years of papersReproduce a paper's experimentCompare forecast errors across modelsRun a survival analysisBuild an epidemic transmission modelSolve a math modeling problemTest the difference between two groupsClean raw dataFind research gaps
Review the last three years of papersReproduce a paper's experimentCompare forecast errors across modelsRun a survival analysisBuild an epidemic transmission modelSolve a math modeling problemTest the difference between two groupsClean raw dataFind research gaps

Close the page, the task keeps running

Code and data run in the cloud. Come back any time to see the results.

My tasks
  • Load forecasting model comparisonRunning2:14:00
  • Epidemic transmission modelQueued0
  • Survival analysisDone0

Every step can be replayed

Searches, runs and edits are recorded in order inside the task.

  • Search papers4.2s
  • Read LOAD_DATA2.8s
  • Run RUN_EXPERIMENT18s
  • Write ERROR_COMPARISON6.1s
Run log
[10:24:17] Searching papers
[10:24:20] Found 12 relevant papers
[10:24:38] Reading LOAD_DATA
[10:25:14] Wrote ERROR_COMPARISON

Everything is kept together

Reports, code, data and figures stay in the same task.

Result files
  • RESEARCH_REPORTOpen
  • RUN_EXPERIMENTOpen
  • LOAD_DATAOpen
  • ERROR_COMPARISONOpen

  1. 1
    Set the research goal

    Describe the question in plain words and attach papers, data or code.

  2. 2
    Search and read papers

    Search, screen relevant work and extract methods, datasets and findings.

  3. 3
    Write and run code

    Run experiments and scientific simulations in the cloud environment.

  4. 4
    Analyze and check results

    Run statistical tests, make figures and check that data and conclusions agree.

  5. 5
    Deliver the results

    Report, code, data and figures are saved together and stay editable.

Task log
Example task · keeps running after you close the page
10:00
Set the research goal
  • Upload the question and LOAD_DATA1m
10:02
Search and read papers
  • Search 4 databases6m
  • Screen 186 → 24 papers12m
10:31
Write and run code
  • Write RUN_EXPERIMENT3m
  • Run LSTM and Transformer41m
11:48
Analyze and check results
  • Paired t-test p = 0.00318m
  • Write ERROR_COMPARISON6m
12:14
Deliver the results
RESEARCH_REPORT
Research report
Open

Search and read papers

Search papers, screen relevant work and extract methods, datasets and findings.

Research task / Load forecasting model comparison · Example
electricity load forecasting transformer lstmScreened 24 / 186
  • Informer: Beyond Efficient Transformer for Long Sequence Time-Series ForecastingAAAI · 2021
  • Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series ForecastingNeurIPS · 2021
  • A Time Series is Worth 64 Words: Long-term Forecasting with TransformersICLR · 2023
  • Are Transformers Effective for Time Series Forecasting?AAAI · 2023
Extracted
PaperMethodFinding
InformerProbSparse self-attentionLower long-horizon error than LSTM baselines
AutoformerDecomposition + auto-correlationLower long-term error than Informer
A Time Series is Worth 64 WordsPatching + channel independenceLower error than earlier Transformers
Are Transformers Effective for Time Series Forecasting?Single linear layerA linear model can beat many Transformers

Run code on a dedicated cloud computer

Each task has its own cloud computer that keeps the code, data and results.

Research task / Load forecasting model comparison · Example
RUN_EXPERIMENTLOAD_DATA
1import torch
2import torch.nn as nn
3
4class LoadForecaster(nn.Module):
5    def __init__(self, input_dim, hidden_dim=128):
6        super().__init__()
7        self.lstm = nn.LSTM(input_dim, hidden_dim, batch_first=True)
8        self.fc = nn.Linear(hidden_dim, 1)
9
10    def forward(self, x):
11        out, _ = self.lstm(x)
12        return self.fc(out[:, -1, :])
13
14for epoch in range(50):
15    loss = train(model, loader)
TerminalCloud computer running
Epoch 36/50 loss 0.0276 val_loss 0.0326
Epoch 37/50 loss 0.0269 val_loss 0.0319
Epoch 38/50 loss 0.0262 val_loss 0.0312

Connect your own server

When you need GPUs or a cluster, connect your server over SSH. Experiments run there and results sync back to the task.

Research task / Load forecasting model comparison · Example
Server connection
Your computer
Scientify
Your server
Connection log
$ ssh user@lab-server
Connected to lab-server
Submitted job RUN_EXPERIMENT
Job running · GPU 0
Synced ERROR_COMPARISON

Researchers

Hand literature reviews, paper reproduction and comparison experiments to Scientify, and spend your time on judgment and design.

Example request
Reproduce this paper's experiment and compare it with the reported results.
Comparing electricity load forecasting models
1. Background
2. Methods
3. Results
4. Conclusion

Graduate students

Proposal research, experiments and paper figures all happen in one task.

Example request
Review the last three years of papers in this area, build a comparison table and find research gaps.
PaperYearMethod
Informer2021ProbSparse self-attention
Autoformer2021Decomposition + auto-correlation
A Time Series is Worth 64 Words2023Patching + channel independence
Are Transformers Effective for Time Series Forecasting?2023Single linear layer

Math modeling teams

Modeling, solving, plotting and the first draft happen in one environment, so contest time goes into ideas.

Example request
Build a forecasting model for this problem and produce code, figures and a first draft.
One-week load forecastActual load · LSTM · Transformer
RUN_EXPERIMENTERROR_COMPARISON

1 credit covers $1 of official API usage.

Starter
$19/ month
60 credits

For occasional literature searches and small analyses

Monthly credits
60
Valid for
1 month
Choose Starter
Researcher
$29/ month
100 credits

For ongoing papers, data analysis and experiments

Monthly credits
100
Valid for
1 month
Choose Researcher
Pro
$99/ month
350 credits

For frequent research tasks and long computations

Monthly credits
350
Valid for
1 month
Choose Pro

New accounts get a free cloud computer and $5 worth of credits.Billing details

Start with one research question

New accounts get a free cloud computer and $5 worth of credits.

Open workspace
Result files
  • RESEARCH_REPORTOpen
  • RUN_EXPERIMENTOpen
  • LOAD_DATAOpen
  • ERROR_COMPARISONOpen