Byte
Join Us
Recruitment Task • ML Research

ML Research Tasks

ML Research / Applied ML

Choose two tasks.

Pick any two of the four tasks across ML Research and Applied ML. Your year determines the required difficulty combination:

  1. First year Any two tasks

    Choose any two of the four tasks. You can pick both from the same track, for example Basic and Advanced Applied ML, or mix tracks and difficulty levels.

  2. Second year At least one Advanced

    Choose any two of the four tasks across both tracks. At least one of your choices must be Advanced.

  3. Third year Both Advanced

    Choose the Advanced task from ML Research and the Advanced task from Applied ML.

Problem Statements

Read through the problem statements below and follow the requirements for your chosen task.

// ACCESS LOCKED //

Problem Statements Locked

Tasks for ML Research will be revealed on 25 SEP 2026 • 15:00 IST.

00// DAYS //
00// HOURS //
00// MINS //
00// SECS //

Review the submission instructions while you wait for the tasks to unlock.

3 Core Requirements

Paper Craft

Choose exactly one paper from the approved list, read it in full, and complete every section of the Paper Craft Reading Template in your own words. You will discuss the paper in your interview, including its claim, surprising findings, and structural choices.

Concept & Inspiration

The template is meant to capture your own reading. Do not use an LLM to fill it in blindly. We want to hear your understanding and reasoning.

Core Requirements

  • Choose exactly one paper from the options below or select a published paper of your own choice.
  • Read the selected paper in full and complete every section of the Paper Craft Reading Template in your own words.
  • Be ready to explain the paper’s central claim, what surprised you, and why the authors made their key structural choices.

Paper Craft Reading Template

The template is available now. You can fill it in.

Paper options: choose exactly one

  • General: An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
  • General: MiniLLM: On-Policy Distillation of Large Language Models
  • General: DeepSeek-V3 Technical Report
  • General: GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints
  • General: Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • RL Domain: Policy Gradient Algorithms, Trust Region Policy Optimization
  • RL Domain: Policy Gradient Algorithms, DeepSeek R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
6 Core Requirements

The Falsification Challenge

Act as a skeptical researcher and investigate the claim that neural networks preferentially learn simpler predictive features. Design a small, controlled study that could falsify, weaken, or qualify the claim, not simply reproduce it. The main deliverable is a rigorous experimental research proposal; running experiments is optional and earns additional credit when it meaningfully strengthens the investigation.

Concept & Inspiration

Ask when simplicity bias might fail, what “simple” means, and whether observed preferences come from feature simplicity, data correlations, architecture, or optimization. Conclusions should match the evidence; you are not expected to disprove the claim.

Core Requirements

  • State a specific research question, a falsifiable hypothesis, an appropriate null hypothesis, and at least one alternative explanation.
  • Propose an implementable experiment suite: describe the dataset or synthetic data, controlled features and variables, what changes and what stays fixed, model architecture, training procedure, and evaluation.
  • Include controls and baselines that distinguish simplicity from competing explanations such as predictive strength or ease of optimization.
  • Discuss multiple possible outcomes and their interpretations, plus limitations, confounders, and what the study cannot establish.
  • Implementation is optional. If feasible, a small, carefully controlled experiment is encouraged; do not add complexity just to produce graphs.
  • Write the proposal and any README yourself. Explain your reasoning and keep conclusions proportional to the evidence.

Full challenge brief

The full research brief includes background and recommended literature, detailed study-design guidance, expected-outcome examples, a suggested proposal structure, and the evaluation criteria.

Submission Protocol

01
Public GitHub Repository

Maintain clean, chronological commits reflecting your build progress. Avoid single-dump commits.

02
Comprehensive README

Document architecture diagrams, setup instructions, trade-offs made, and security measures.

03
Hosted Demo / Walkthrough

Provide a live link (Vercel/Netlify), shareable APK/Expo link, or screen recording walkthrough.

04
Official Form Submission

Submit before 16 Oct 2026 at 11:59 PM IST. Shortlisted candidates will be invited for live walkthroughs.

Explore Other Department Tasks

App DevelopmentBuild apps people actually use.
Web DevelopmentShip it. Then ship it faster.
Applied MLBuild systems that reason and take action.
CADDesign it. Assemble it. Make it work.
ElectronicsSense, respond, and build reliable circuits.
ROSRobot Operating System
CybersecurityBreak things. Then figure out why.
Graphic DesignMake BYTE impossible to scroll past.
Video EditingTurn raw footage into a story.
OutreachRead the room. Make the ask.