Public articles linked to the same research event.
arXiv The work introduces APDMem (Agent-controlled Progressive Disclosure Memory), which organizes conversation history into four progressively detailed layers—thematic summaries, personalized key facts, turn-level evidence notes, and raw messages—has a controller read high-level summaries first and drill into finer evidence only when needed, and uses a note synthesizer to turn retrieved evidence into a query-focused structure that consolidates facts, orders events, and flags contradictions, achieving strong long-context memory reasoning on LongMemEval while accessing only 8% of the total conversations.
The work introduces APDMem (Agent-controlled Progressive Disclosure Memory), which organizes conversation history into four progressively detailed layers—thematic summaries, personalized key facts, turn-level evidence notes, and raw messages—has a controller read high-level summaries first and drill into finer evidence only when needed, and uses a note synthesizer to turn retrieved evidence into a query-focused structure that consolidates facts, orders events, and flags contradictions, achieving strong long-context memory reasoning on LongMemEval while accessing only 8% of the total conversations.
The work introduces APDMem (Agent-controlled Progressive Disclosure Memory), which organizes conversation history into four progressively detailed layers—thematic summaries, personalized key facts, turn-level evidence notes, and raw messages—has a controller read high-level summaries first and drill into finer evidence only when needed, and uses a note synthesizer to turn retrieved evidence into a query-focused structure that consolidates facts, orders events, and flags contradictions, achieving strong long-context memory reasoning on LongMemEval while accessing only 8% of the total conversations.
The work introduces APDMem (Agent-controlled Progressive Disclosure Memory), which organizes conversation history into four progressively detailed layers—thematic summaries, personalized key facts, turn-level evidence notes, and raw messages—has a controller read high-level summaries first and drill into finer evidence only when needed, and uses a note synthesizer to turn retrieved evidence into a query-focused structure that consolidates facts, orders events, and flags contradictions, achieving strong long-context memory reasoning on LongMemEval while accessing only 8% of the total conversations.