When should an agent ask for help?

I’m Vicky Feliren, an applied scientist interested in decisions under uncertainty. I study when an agent should act, gather more evidence, or ask for help.

Read the research View the CV ↗

Vicky Feliren

8 Publications 3 journal articles, 4 conference papers, and 1 preprint, including 3 as first author.

5+ years Industry and freelance ML Work across employed roles and freelance projects, including models for finance and city services.

1 Issued patent A method that passed formal examination and became an issued patent.

12 Awards & scholarships Including the Asia-Pacific regional award at the Global South AI Safety Hackathon.

100+ Researchers co-led with Experience coordinating research across a large international group.

1000+ Students taught & judged Experience explaining difficult ideas to people seeing them for the first time.

Background

At Jakarta Smart City, I forecast municipal waste. Later, at GDP Labs, I built banking models. At Monash, I studied flood and mining maps from satellite images. Alongside this work, I help build Southeast Asian datasets with SEACrowd.

These settings shaped my interest in how uncertainty should influence a decision.

View the full CV →

Research interests

Each choice an agent makes changes what it encounters next. Therefore, I want to study whether uncertainty estimates remain reliable across a whole task.

I’m interested in how changes in inputs, models, and error costs affect these decisions. I also want to test whether internal model signals help identify when an answer needs review.

See my work and next questions →

Each choice changes the next decision.

A navigation agent reaches an unfamiliar junction.

• Act

  The agent enters a corridor. Its next decision depends on what it sees there.

• Gather evidence

  The agent inspects its surroundings. It gains more observations before choosing a corridor.

• Ask for help

  The agent pauses for guidance. Its next decision uses the additional instruction.